155 Commits

Include renames

Author	SHA1	Message	Date
mudler-agentandEttore Di Giacinto	4899d28e8a	feat(chats): server-side chats with steering, sub-agents and compaction (#12918) ... * Find the Sandbox guest image in the executable, not its build info The Sandbox smoke checked that the packaged executable carries the guest image by looking for -X main.guestImage in go version -m. The package is built with -trimpath, and with -trimpath Go leaves -ldflags out of the build info, so the check failed on every gvisor run, although the image was linked in. Look for the reference in the executable instead: the value of SANDBOX_GUEST_IMAGE when it is set, any digest reference otherwise. package.sh had the same check for an executable built elsewhere; it now looks for its own reference the same way. Assisted-by: Claude:claude-opus-5-5 * Answer a network card the limit refuses with one precise reason A card that waited while the administrator's limit went down was refused on the first click with 400, and every later click was a bare 409, so a stale card looked like a conflict loop. A refused card now answers 400 egress_above_ceiling every time, the same answer sent twice is one answer (204), and another answer to an answered card is 409 already_answered. The chat view marks a card the limit cannot admit with above_ceiling. When the limit is off, or below the chat's preset, the note of the turn and the block line of the proxy now say that the server's setting blocks the network and no approval changes it, so the model reports it instead of waiting or retrying. Assisted-by: Claude:claude-sonnet-5-5 * Explain a blocked Sandbox network instead of offering dead buttons A card the limit cannot admit shows why network access is off or limited, and who can change it, with a link to Settings > Tools > Sandbox network for administrators. The buttons of a card disable while an answer is sent, and a failed answer reads the card again. The Network row says why its choices are capped. Settings opens a group from ?group=. Strings are translated in all nine locales and the chat and add-on docs describe the new behaviour. Assisted-by: Claude:claude-sonnet-5-5 * Add an always-allowed hosts list to the Sandbox network The extra registries setting only widens the package registries, so a one-off host needs the approval card in every chat. The new admin setting chats_egress_allowed_hosts lists exact hosts (optionally host:port to pin a port) that a chat reaches whenever its effective preset is registries or all. The list sits in the same decision as the presets: the ceiling still caps it, the ports setting still applies, and a private, loopback, link-local or metadata address still needs the private switch and an administrator's rule or answer. Each use is audited with the reason allowed_host. Hosts are validated like the extra registries (lower case, punycode, no scheme, path or pattern, at most 64) and an IP literal is refused unless the private switch is on. get_chat_egress and the chat egress view return the list; no MCP tool writes it, as for the extra registries. Assisted-by: Claude:claude-sonnet-5-5 * Let users delete from the Trash for good A delete only set a tombstone, so the Trash grew without end and nothing could free the space of a deleted file. The workspace stores gain a Purger: it erases the deleted nodes of some change sets, or of a whole folder, with their journal rows, the change sets that nothing is left in, and the blobs that no row refers to. Both stores implement it and pass the same contract specs. A blob is removed after a grace period, and only if it is still unused then, so an upload of the same bytes in flight keeps them. DELETE /api/chats/files/trash empties the user's Trash and DELETE /api/chats/files/trash/{rev} deletes one entry. Both are scoped to the caller's home. The Trash listing now gives files, folders and bytes per entry and for the whole Trash. The file tools of a chat cannot reach any of this. Assisted-by: Claude:claude-sonnet-5-5 * Add Empty Trash and Delete permanently to the Files page The Trash tab shows what it holds, with an Empty Trash button that is off while the Trash is empty and a Delete permanently button on each item. Both ask in the page first and say how many files and folders and how much space they remove. Texts are translated in all nine locales. Assisted-by: Claude:claude-sonnet-5-5 * Fall back to Brave Search and Bing when DuckDuckGo asks for a captcha DuckDuckGo answers some servers with HTTP 202 and a bot check (the "anomaly" page) instead of results. The parser read that page as "no results it could read", so web_search failed for every query. The keyless search is now a chain: DuckDuckGo, then the result page of Brave Search, then the result page of Bing. Each page has its own tolerant parser (title, URL, snippet; Bing's /ck/a links and DDG's uddg links are unwrapped) with saved fixtures. A bot check is a typed error (ErrChallenged), and a 403 or 429 is ErrBlocked. Both wrap ErrProvider. A Fallback with a cool-down skips a challenged or blocked provider for a few minutes, so each search does not wait for it first; when every provider is cooling down, all are asked again. Fallback.SearchAttempts and TestEach report what each provider did (ok, no_results, challenged, blocked, timeout, error, skipped). The error a model reads now lists what was tried and what each provider did, then where an administrator fixes it. safehttp keeps the status code of an answer (Response.StatusCode) and of a refusal (StatusError, still ErrStatus for errors.Is). Assisted-by: Claude:claude-opus-5-5 * Use the keyless search chain in the chats and test each provider The built-in web_search now searches with DuckDuckGo, Brave Search and Bing in turn when no SearXNG instance is set. A new runtime setting, chats_web_search_providers, sets the order (searxng, duckduckgo, brave_html, bing_html) and leaves a provider out. Empty keeps the old choice: SearXNG alone when an instance is set, so its queries do not reach the public engines unless the administrator lists them. POST /api/chats/admin/web/test asks every provider of the order and returns what each did in "providers" (ok, no_results, challenged, blocked, timeout or error, with the reason), next to the first answer. Settings > Tools has a Test search on the web_search row that shows it, and the provider list is checked on the page and by the server. The cool-down of a challenged provider is kept per frontend, like the cache and the limits; web.md says why. Docs: web.md (providers, chain, bot checks, privacy, how to fix it with SearXNG or the Web add-on) and runtime-settings.md. Strings in the nine locales. Swagger regenerated. Assisted-by: Claude:claude-opus-5-5 * Try the mirror library alone in the reinstall recovery spec The spec adds a second library with the same URL and expects an install from it to keep the retained data apart. Since two enabled libraries that list one add-on are a conflict, the reinstall from the original library then failed with library_conflict, and the install from the mirror failed for that reason instead of the trusted identity. Switch the original library off while the mirror is tried, check that the mirror is refused for its identity and not for a conflict, then switch the mirror off and the original on for the reinstall. Assisted-by: Claude:claude-opus-5-5 * Pause a chat turn at the step limit and offer Continue A chat turn ran at most 10 rounds of tool calls, the default of the harness, because nothing set a limit. A multi-step task (read some files, run commands, write a result) stopped half way, and the person saw an answer that looked final. The limit is now the runtime setting chats_max_iterations (1 to 500, default 40, Settings > Tools); a model's agent.max_iterations still wins. When the tool loop reaches it, the harness sends a paused event with the limit, and the turn ends with last_turn "paused". A paused turn counts as finished: waiting messages start the next turn and a sub-agent reports back. The status line says "Paused, after 40 steps" and has a Continue button, which sends a message that asks the model to go on. A claim that found no waiting message released the chat as "done", which hid a pause. It now keeps the status the chat had. Assisted-by: Claude:claude-opus-5-5 * Edit the always-allowed Sandbox hosts in Settings > Tools Add a list editor for chats_egress_allowed_hosts next to the extra package registries: the hosts are tags with a remove button, an input adds one on Enter, comma or Add (a pasted list adds them all), and the reason an entry is refused shows under the input, with the count out of 64. The same checks stop Apply on a bad entry, and the server still has the last word. The chat's Network panel lists the hosts read only and the row says how many an administrator always allows while the chat's network is on. The help text of both settings now says how they differ. Strings are translated in all nine locales. Assisted-by: Claude:claude-sonnet-5-5 * Document the always-allowed Sandbox hosts Describe chats_egress_allowed_hosts in the runtime settings page, the network section of the add-ons page (including the allowed_host audit reason and how the list differs from extra registries) and the chats page. Assisted-by: Claude:claude-sonnet-5-5 * Let a chat edit, pause and delete the person's triggers A chat could propose a trigger and list them, but not change one: the Triggers page and the API could. list_triggers now gives each trigger a stable ref (not its id) and shows the prompt of one trigger only when asked. update_trigger and delete_trigger store a change record in the trigger table: an edit or a delete waits for Apply or Delete on a card, a pause or a resume applies at once with a receipt. An applied change can be undone for a day; a delete restores the trigger with its id. Change records are rows in the database, so every frontend shows and answers the same card. The new routes under /api/chats/triggers/changes serve the card and are gated by the triggers feature. Assisted-by: Claude:claude-opus-5-5 * Show the trigger changes a chat asks for as cards An edit lists each changed field before and after with Apply and Dismiss, a delete shows the trigger and its last run with Delete and Keep, and an applied change is a receipt with Undo. The cards sit below the message text like the proposal card, and read their state from the server so a reload shows the same answer. Assisted-by: Claude:claude-opus-5-5 * Add a Test search button to the web tools settings POST /api/chats/admin/web/test existed, but no page called it and the docs showed only curl. The web_search row of Settings > Tools now has a Test search button. It runs one search with the applied settings and shows the provider with its first results, or the error a chat would read. Assisted-by: Claude:claude-opus-5-5 * Describe undo per tool call in the chats docs The docs said the checkpoint line of a turn carries the change set of the turn. A turn checkpoint is taken only by a chat with a VM sandbox, which the service never sets up, so that rev is always 0. What ships is one change set per file tool call, with a files_change card on its tool line. The docs now say so and how to undo a whole turn. Grouping a turn's changes into one change set in the file store is left for later: it touches the file store's revisions and the undo of shared folders. Assisted-by: Claude:claude-opus-5-5 * Mount the skill folders read-only for the built-in bash The built-in pdf skill ships scripts, but the commands saw only the home and the shared folders, so the model could read the scripts and not run them. The built-in bash now mounts the three skill layers read-only, at the names the file tools already use: shared-skills, shared-builtin-skills and shared-addon-skills. A shared folder named skills moves to "shared-skills 2", so the two never collide. A read-only mount with subfolders made every command report its folders as changes the mount refused, because the host backend lists every folder of its work tree. A folder alone is no change: both stores now refuse only files, whiteouts and opaque folders there. The Sandbox add-on mounts only the home; the docs say to copy a script there first. Assisted-by: Claude:claude-opus-5-5 * Find the add-on that runs commands by what it shadows The chats knew the add-on that runs commands from a per-process cache, filled with the Sandbox's own tool name the first time a chat listed its tools. Another add-on that replaces bash did not count, so skills that need commands showed as unavailable. The admin network view read the same cache: after a restart it said no Sandbox serves until someone chatted, each frontend of a cluster could answer differently, and a removed add-on stayed listed. The tool that runs commands is now the one that shadows bash, as the spec asks. sandboxRunTool and the cache are gone. The stand-in of an add-on that cannot work carries the built-in tools it replaces, and the admin view asks the add-on host which add-on serves bash now: the pinned one or the oldest running install, and none when the pinned one is stopped. Assisted-by: Claude:claude-opus-5-5 * Show international host names in punycode in the network checks The card-spoofing defence of the sandbox network asks for host names in their ASCII form, but CheckHost passed Unicode names through. A look-alike name reached the consent card as text that reads like the real site. CheckHost now converts an international name with the IDNA lookup profile and refuses one with no valid ASCII form. The conversion runs before the address checks, so full-width digits that spell a loopback address are refused as loopback. Assisted-by: Claude:claude-opus-5-5 * Keep the seeded Studio and Talk pins until the add-ons are installed The upgrade pins Studio and Talk for existing accounts, but both are add-ons that an administrator installs later. The sweep marked a pin whose add-on is not installed as missing and deleted it after 30 days, so an administrator who installed Studio late found the seeded pins gone. The sweep now records the first time a seeded add-on is present, in the table of seed runs or the pins file. A seeded pin keeps no clock until then, and a mark that an older sweep set is cleared. After the first install the pin follows the retention rule as before. Assisted-by: Claude:claude-opus-5-5 * Fix two small labels of the Skills and Backends pages On the Skills page, the skill of an add-on that is not chosen said "Replaced by <add-on>", the label of a built-in skill that an add-on serves. It now says "Not active: <add-on> provides it", as the add-on skills spec asks. On the Backends page, Clear filters on an empty Installed search left the library filter set, so the list stayed empty. It now clears the library filter too. Assisted-by: Claude:claude-opus-5-5 * Tell the model when no network card can appear Each turn told the model that a blocked host asks the user. No card appears when the ceiling is off or no wider than the chat's preset, in an unattended run or in a sub-agent, so the model waited for an answer that could not come. The line now says that a blocked host is refused and nothing asks the user in those cases, so the model names the host it needs in its answer. Assisted-by: Claude:claude-opus-5-5 * Translate six Backends and preview labels in every locale check, reinstallBackend, deleteBackend and filterLabel of the Backends page and the label and failed text of the file preview showed in English in the eight other locales. They now have translations, and the i18n baseline goes down to match. Also fix a comment of the home composer that still named the removed Canvas chip. Assisted-by: Claude:claude-opus-5-5 * List the admin routes that have no assistant tool, with the reason The admin oversight routes of saved agents, triggers and connectors had no MCP tool and no recorded reason, and the other deliberate skips lived only in code comments. coverage_test.go now keeps them in adminRoutesWithoutTool with a reason each, and a spec checks that no route there is also wrapped by a tool. The contributor checklist names the map. The tool catalog of the assistant docs said it matched the real one but missed most tools added since. It now lists every tool, read-only and mutating, and the actions that stay in the web interface. Assisted-by: Claude:claude-opus-5-5 * Close the doc and registry gaps of the chats and add-ons - /api/agent/* answers 404 with the chats_off reason when the chats are off, as /api/chats/* and /api/workspace/* do. - The add-ons docs get an upgrade section: Studio and Talk are add-ons, nothing is preinstalled, and users need the addons feature, which the upgrade does not grant. It says how an administrator installs them (also from a private library) and grants the feature, and carries a paragraph for the release notes. - The migration page no longer lists fork, edit, the chat settings and the token counters as removed. - Runtime settings document chats_vision_model, chats_transcription_model and the skill keys of chats_tool_providers. - The docs show POST /api/chats/files/folders, GET /api/chats/files/history and the allow route of a connector sender. - The in-tree specs get amendments where the shipped design differs. - A route test of the tagged suite still set the removed Runs field. Assisted-by: Claude:claude-opus-5-5 * Look for the kit empty state in the backends regression spec The Backends page now draws the shared EmptyState of the kit (ui-empty), not the old dk-empty block, so the regression spec of the gallery suite found no empty state after a search with no match. Check the kit's no-results form and its title. Assisted-by: Claude:claude-opus-5-5 * Ignore a features request that outlives its page in the add-on stubs stubAddonRoutes answers /api/features by fetching the real flags and turning add-ons on. A spec that ends while the page polls, like the one that checks that a viewer without the feature never reads the list, left that fetch in flight, and Playwright failed the spec with "route.fetch: Target page, context or browser has been closed" after its checks had passed. Abort the request when the fetch fails because the page is gone. Assisted-by: Claude:claude-opus-5-5 * Bring the gallery-ui case inventory up to date The suite lists the node unit cases of addons.test.mjs and the Playwright specs of addons-gallery.spec.js. The inventory missed the unit cases added since it was written, and one spec was renamed when the legacy switch came in ("hub: the Add-ons tab appears only for an account that holds the feature" is now "hub: under the legacy switch the Add-ons tab appears ..."). The list comes from a local run of both parts against the UI test server. Assisted-by: Claude:claude-opus-5-5 * Read spreadsheets, documents and slides in read_document read_document said "file.xlsx has no text layer; it may be a scan" for workbooks that some Python libraries write. Their workbook gives the sheet as an absolute part name (/xl/worksheets/sheet1.xml) and the cells as inline strings. The text extractor of the harness joined "xl/" with that name, found no sheet, and returned no text; the tool then took any empty result for a scanned PDF. The new package core/services/studio/extract has one entry point, Extract(ctx, name, data), that returns the text and the structure it read, for read_document now and for a file indexer later. Office files go through the files preview reader, which resolves both kinds of part names and reads inline strings: - a workbook comes sheet by sheet, rows as tab-separated lines, up to 2000 rows and 50 columns a sheet, with a note for a cut sheet; - a Word document gives headings, list items, paragraphs and tables; - a presentation gives each slide with its title, text and notes. PDF and EPUB keep the harness extractor. Only a PDF with no text is reported as a probable scan; an empty Office file is reported as empty. The preview reader now keeps a Word table as one block with its rows, and the preview pane draws it as a table. Assisted-by: Claude:claude-opus-5-5 * Add the missing explorer, library and tools translations Eight locales had no explorer.json, library.json or tools.json. The browser asks the server for every namespace of the language it shows, so an Italian page that used the library answered 404 for /locales/it/library.json on every load and fell back to English. The 24 files now exist, translated, with every key of English. The locale gate counted a missing file as missing keys, which the baseline allowed. It now fails on a missing namespace file whatever the baseline says, and the baseline is 24 entries shorter. Assisted-by: Claude:claude-opus-5-5 * Back off while the add-on service is unavailable A page that followed an add-on upgrade read the operation every two seconds, up to 300 times. While the add-on service restarted, each read waited for its timeout and answered 503, so the log of one session had 48 failed dispatches in a row, and the page only showed the upgrade as running. The Sandbox state of a chat's Tools menu read once and, on a 503, said nothing more. The reads of an operation now back off while the service does not answer: 2, 4, 8, then up to 30 seconds. The upgrade entry says that it waits for the add-on service and when it checks again, in Activity and in the details, which also get a Check now button. The Tools menu says that the Sandbox is starting and reads its state again with the same back-off, kept for the page, so opening another chat does not ask again before the wait is over. Assisted-by: Claude:claude-opus-5-5 * Start a chat with its switches and list them before it exists The composer of a new chat could not show the Tools and Skills menus, because the lists and the switches belong to a chat. A chat started from Home therefore always began with every tool and skill on. - POST /api/chats takes switches: the add-ons, MCP servers and connectors, built-in groups and skills that are off, and the opt-in groups that are on. They are checked as the PUT endpoints check them and stored with the new record, so the chat never exists without them; a bad entry refuses the request and creates nothing. The SQL store now writes these lists on create too. - GET /api/chats/draft/tools and /api/chats/draft/addons list what a chat on a model, with a saved agent, could use, every switch in its default state. They refuse what create refuses: an unknown or forbidden model, an agent without the agents feature, another user's agent. Document both in the chats page and regenerate the swagger. Assisted-by: Claude:claude-opus-5-5 * Give Home the composer of the chat page Home drew a smaller message box than the chat page: no files, no agent, no Tools or Skills menus, no paste to attach, no rich model list, and its text was lost on a reload. A chat started there could not be set up before its first message. Move the composer wiring out of the chat page into one place that both pages use: - useChatComposer holds the draft (text and files), the model and the saved agent of a new chat, the chat models, the switches and what attaching and pasting do. - ChatComposer draws the model list (fit, vision, memory bar), the agent picker, the Tools and Skills menus, the attach buttons, dictation, the add-on slot and the vision notice over HomeComposer. - useChatSwitches has a draft mode: before the chat exists it lists from the draft endpoints, a switch changes only the screen, and the choices go with the create request. They outlive a change of model or agent. Home now creates the chat with the model, the agent and the drafted switches, sends the first message with its files and opens the chat. When the message cannot go, the new chat is removed and the text and files stay on Home. Its text is kept in the browser across a reload. Home also gets the shortcuts row, the notices of the chat page and the line that says whether the model is loaded. Its own slash actions, widgets, examples and first-run steps stay. The empty chat page gets the same draft menus. On a phone the chips of both wrap to a second row, and the Tools and agent menus open as a sheet from the bottom, as the skills picker does. Assisted-by: Claude:claude-opus-5-5 * Move add-on settings to the add-on's own page The settings.sections slot drew every add-on's settings at the end of the global Settings page, and the Sandbox and Web sections were unstyled: the add-on UIs used class names the host does not define (btn, btn-primary) and bare inputs, the SDK offered no form primitives, its muted token named a property that does not exist, and the slot root was an inline-flex span that laid a form out in a row. An add-on's settings sections now render on the Settings tab of its details on the Add-ons page (/app/addons?detail=<id>&view=settings), narrowed to that add-on, for administrators only. An add-on page gets a host Settings link to that tab. The Settings page no longer draws the slot: an Add-ons group, shown when an add-on has settings, lists those add-ons with a link each. The old #ad-section-<addon>-<id> anchor redirects to the tab once the apps list is read. Server settings of an add-on's feature stay on the Settings page; the tab shows them under "Server settings for this add-on", with a read-only summary of the Sandbox network and links to ?group=tools#settings-egress and #settings-builtin-tools. The SDK gets a settings kit in sdk.ui (Section, Field, Select, TextInput, Switch, Choice, StatusChip, Callout, List/ListRow, Code, Actions, Table, ExternalLink), drawn with host classes and tokens, and more theme tokens. The Sandbox and Web sections are rebuilt with it: backend rows with a status chip and the raw reason collapsed in a code block, a backend choice with one line per option, and the AppArmor and user-namespace hint (with a docs link) only for the gVisor "exit status 128 ... cannot read client sync file" failure. The Settings page showed "Changed, default 30, Reset" next to a 0 for the audit retention, and the same for the Sandbox network limit, preset and ports: the server stores 0 or an empty value for a setting never set. Those fields now read that value as their default, and the retention input has its unit and bounds. Assisted-by: Claude:claude-opus-5-5 * Adapt the add-on runtime specs to the new navigation Three specs of the esm-ui suite still assumed the old shell: - the Add-ons tab that lists the pages add-ons declare is part of the legacy layout, so the navigation spec found no hub subnav; it now sets the legacy switch on the server's flags for that page; - the sidebar has no "Chat" link any more, so the unsaved-edits spec could not leave the page; it leaves through Home in the primary navigation; - the shell keeps a notification stream open, so the network never goes idle and the spec for a member without the feature waited out its timeout; it waits for the server's switches, the account and the shell instead. Assisted-by: Claude:claude-opus-5-5 * Name the legacy Add-ons tab unit case as the esm unit tests do The esm-ui inventory listed "unit: registry: the Add-ons tab lists page links only when the registry has usable add-ons", but the unit test is named "the legacy Add-ons tab lists ...", so the suite could never match its inventory. Use the name the test prints. Assisted-by: Claude:claude-opus-5-5 * Say on the Collections page when the embedding model is missing At startup the server opened a collection and its first embedding call got 404 from POST /embeddings. The collections embed through this server's own API with collections_embedding_model, the default granite-embedding-107m-multilingual, and that model was not installed. Nothing told the user: uploads and searches just failed. GET /api/collections now also returns embedding_model and embedding_model_installed (a loaded configuration of that name, or a model file of that name). When it is missing, the Collections page says "No embedding model installed", names the model, and gives an administrator a button that opens the Models page with a search for it. Assisted-by: Claude:claude-opus-5-5 * Report modified times and the upload limit in folder listings The Files page could not sort by date or show when a file changed: a folder listing had no time. It also learned the upload limit only after it had sent the whole file. The default server body limit is 15 MB, lower than the 256 MiB a workspace file may have, so a large upload failed late and without a clear reason. Each listed entry now has "modified", the time of the change set of its revision. Both workspace stores fill it in List: the JSON store from its in-memory change sets, the database store with one query per batch of revisions. The listing also has "max_file_size", the smaller of the workspace limit and the server body limit, so a client can refuse a larger file before it sends any byte. The store contract suite checks the new time against the history, and the route tests check both fields. Assisted-by: Claude:claude-opus-5-5 * Add a sortable file list with thumbnails for the Files page The building blocks of the new file browser, used by the page in a later change: - fileBrowser.js: the kind of an entry from its extension, the order (folders first, natural and case-insensitive names, then size, modified time or type), the filter of a folder, dates in words with the full date for a tooltip, the Shift range, the arrow moves of a list and of a grid, type-to-find, names that do not clash, why a name cannot be used, the name of a pasted screenshot and the view settings read back from storage. - FileList: a grid with sortable column headers or a grid of tiles with image thumbnails, a check box per row, a row menu, inline rename with the reason a name is refused, rows that can be dragged onto folders, and the keyboard: arrows, Shift, Space, Enter, F2, Delete, Ctrl+A, Escape, Backspace and typing a name. - MoveDialog: a folder picker that walks the tree and refuses a move into the moving folder itself. - ActionMenu can open at a point with no trigger (a right click), and its trigger can take another icon and class (a split button). - fileSize also writes gigabytes. Assisted-by: Claude:claude-opus-5-5 * Add an upload queue with progress, cancel and retry An upload of the Files page was one fetch per file in a loop: nothing showed while it ran, one failure stopped the rest, and leaving the page lost it. - uploadQueue.js keeps every upload outside React: queued, sending with the bytes sent and a smoothed rate, done, failed with a reason and whether Try again can help, or canceled. Two files go at once. It lives for the whole app, so opening another folder or page does not stop an upload. - filesApi.uploadWithProgress sends the same PUT with XMLHttpRequest, which reports upload progress and can be aborted. The server already streams the body into the store; nothing changes there. - useUploads reads the queue and asks before the tab closes while a file is on its way. - UploadTray shows the queue at the bottom of the page: the total and the time left, each file with its progress, speed and time left, Cancel, Try again with the reason, and a done state that fades out. A live region tells a screen reader when uploads start, fail and end. - ConflictDialog asks what to do when an upload has the name of a file in the folder: replace it, keep both, or skip it. - dropFiles.js reads dropped folders with their paths. Assisted-by: Claude:claude-opus-5-5 * Rebuild the Files page as a file browser The Files page was a list of links: the store's byte order, the same icon for every file, three icon buttons on every row, no progress on an upload, no selection, and on a phone a preview that opened below the whole list. Browsing: a breadcrumb that folds a deep path, sortable columns (name, modified, type, size) with folders first, a list or a grid of thumbnails, both kept per user in the browser, a filter of the open folder with a search of every file on Enter, type icons, human sizes, dates in words with the full date on hover, and Back and Forward between folders. Uploads go through the upload queue and its tray: the Upload button or its folder menu, drops of files and folders anywhere on the page or onto a folder row or crumb, and paste. A clash asks to replace, keep both or skip; a file over the server limit is refused before it is sent. Selection and actions: click, Ctrl and Shift, check boxes, a selection bar (Download, Move to, Move to Trash; at the bottom of the screen on a phone), a row menu and a right click, inline rename and new folder or file, drag to move, the keyboard, and a toast with Undo for a delete, a move and a rename. States: a skeleton on the first load, an error with Try again, empty states that say how files get here, and a no-match state. The preview uses the side pane: docked on a wide screen, a sheet on a phone, with the same actions as the row menu. Trash, History and the Managed by LocalAI group stay as they were. Every new string is in the nine languages. The page uses design tokens only (seven raw font sizes fewer). The specs cover sorting, the filter, the views, selection and bulk actions, the keyboard, the menu, rename checks, drag to move, drops, paste, clashes, the size limit, upload progress, cancel and retry against the real server, and the error states. Assisted-by: Claude:claude-opus-5-5 * Fix two specs that fail on the integration branch The Describe spec of the egress package checked that the no-card line does not contain "asks the user", but that line says "nothing asks the user". It now checks for the sentence that promises a card. The context meter spec gave the chat a 4k context window. The tool schemas of the chat now take 3k of it, so the turn stopped before the model request. The spec uses a 16k window. Assisted-by: Claude:claude-opus-5-5 * Translate the always-allowed hosts note of the chat network row The three egress keys of the always-allowed hosts were only in English, so the locale check failed for every chats namespace. The workspace namespaces now miss two keys fewer, so their baseline goes down too. Assisted-by: Claude:claude-opus-5-5 * Bring the contracts and host-slots case inventories up to date The contracts suite has eight new specs for the check of a tool that shadows a built-in. The host-slots suite runs the unit case of the file preview slot, which the inventory did not list. Assisted-by: Claude:claude-opus-5-5 * Add the design of the file text index The search of the chats' files reads only small UTF-8 text files, so a PDF or an Office file is found by name only. The design extracts the text when a file is written, keeps it in LocalAI's own store keyed by node and hash, checks access at query time and runs one indexer per cluster under an advisory lock. Assisted-by: Claude:claude-opus-5-5 * Give extracted documents their pages, sheets and slides Extract now says where each page, sheet and slide starts in the text, so a search hit can name its page or sheet. A PDF page starts with a "## Page N" line, as sheets and slides already start with a heading. A PDF also stops at a page limit (200 by default) and at the text limit, which it ignored before, and a panic of the reader is an error. The test fixtures move to extracttest, with a real text PDF and a workbook with shared strings, so other packages can build the same documents. Assisted-by: Claude:claude-opus-5-5 * Add a store for the text of workspace documents textindex keeps one entry per workspace node: the text read from a PDF or Office file, the hash it was read from, its pages or sheets and its status (pending, ready, failed, skipped), plus the cursor of the indexer in the workspace journal. The GORM store keeps a lower-cased copy of the text, so SQL and the Go matcher agree on every script, and searches it with FTS5 and the trigram tokenizer on SQLite or a pg_trgm index on PostgreSQL. Without them it scans the column. The JSON store keeps the metadata in memory and each text in its own file, and scans the texts. One contract suite runs on JSON, SQLite and PostgreSQL. Assisted-by: Claude:claude-opus-5-5 * Index the text of new and changed workspace documents The indexer tails the workspace journal from a cursor kept in the text store. A write of a PDF, Word, Excel or PowerPoint file marks it pending (the same hash only follows the path), a delete removes the entry and a move rewrites the paths below a folder. Then it reads the pending files two at a time, with a timeout, a size cap, a page cap and a text cap, and stores ready, failed or skipped. A panic of the reader fails the file, not the pass. The same content is read once. The first pass starts at revision 0, which indexes the files that were there before, in bounded steps that resume from the cursor. Enqueue marks an upload pending at once and Forget drops the entries of nodes a purge erased. In a cluster each pass runs under a new advisory lock, so one frontend reads. Assisted-by: Claude:claude-opus-5-5 * Search and read the text of documents in the chats' files search_files and GET /api/chats/files/search now also search the text index: after the name and text-file hits come lines of PDF, Word, Excel and PowerPoint files, as path (page 3, line 40): text, with the sheet or slide where a document has them. Each entry is shown only when the caller may read its file now and the file still has the content the text came from. Documents keep up to a third of the hits. When files below the path are still pending, the answer says so instead of "Nothing found.", and the REST result has indexing. read_file returns the text of a PDF or Office file with the usual offset and limit, from the index when it matches the file and read on the spot otherwise. An upload and a write of a chat tool queue the file at once, and emptying the Trash drops the entries of the purged files. Assisted-by: Claude:claude-opus-5-5 * Start the file text index with the files and add its settings With files on, the server opens the text store next to the workspace index (the auth database, or a folder under the data path) and starts the indexer. A chat worker searches the index but runs no indexer. chats_index_files (default on) switches the reading off and chats_index_max_text_bytes (2 MiB by default, 64 KiB to 16 MiB) caps the text kept for one document. Both apply at the next pass, and both are rows of Settings > Tools. Assisted-by: Claude:claude-opus-5-5 * Show the page, sheet or slide of a document hit on the Files page Search all files now lists hits inside PDF and Office files with their page, sheet or slide instead of a line number of the extracted text, and says how many documents are still being indexed. The strings are in all nine locales. Assisted-by: Claude:claude-opus-5-5 * Test finding an uploaded PDF by its words through the routes The PDF has a generated name and is uploaded with the REST upload. The search route first reports it as being indexed, then finds it on its page after a pass of the indexer, and so does the search_files tool of a chat. Another user finds nothing. Assisted-by: Claude:claude-opus-5-5 * Document the search inside documents The chats page says what the index reads, where the text is kept, who finds it, which search engine each database uses, the two settings, the backfill and the single-runner mode in a cluster, and how search_files and read_file answer for documents. The API instructions and the PDF skill say the same in short. Assisted-by: Claude:claude-opus-5-5 * Test the indexer on the JSON workspace store A server without a database keeps the workspace and the text in JSON files. The indexer follows the uploads, a folder move and a delete in that journal as it does in the database one. Assisted-by: Claude:claude-opus-5-5 * Read all of a long document with read_file The index keeps the first chats_index_max_text_bytes of a document, so read_file used to stop where the index stopped. It now uses the index only when it holds the whole text, and reads the document again (up to 16 MiB of text) when the index cut it. Assisted-by: Claude:claude-opus-5-5 * Start bash in the user's home The Sandbox's bash and the built-in bash started in /workspace, the folder that holds the mounts, so a model that ran a script by a relative path failed and a file it saved there was lost. A call now starts in /workspace/home, the folder of the user's files: a relative path is the same file the file tools see. The tool descriptions and the commands block of every turn say where a call starts and that only what is saved there is kept. The built-in bash says that the /workspace paths do not exist on the host and that a shared folder is next to the home. Assisted-by: Claude:claude-opus-5-5 * Name files in tool results by the path the model uses write_file, edit_file and search_files named a file of the home by its path in the home ("/scripts/build.py") while the model had written /workspace/home/scripts/build.py, and a model took the two spellings for two files. The results now give the path the model uses, and move_file says where it moved the file to. The cards and the change notes keep the path in the home. Assisted-by: Claude:claude-opus-5-5 * Name the paths a command's commit did not save The result of a bash call ended with "N file(s) changed in your files meanwhile; the sandbox's version was kept as a copy" for every path the commit refused, also for a file written outside the user's files, where no copy is kept. It now has one line a path, with the path as the guest sees it, the reason, and the path of the copy when the store kept one, so the model and the user can act on it. The host returns the paths and the copy to the Sandbox add-on relative to the user's files, as its mounts are. Assisted-by: Claude:claude-opus-5-5 * Show the files a command saved under its step A document that a script wrote with bash reached the user's files, but the chat showed nothing: only the file tools put a file on their step, and the Sandbox committed for no chat, so the chat's file list, the chips under the answer and the preview pane never saw the file. The host now commits a Sandbox call for the chat that ran it (the action resource of the request names the chat). After each bash call, of the Sandbox or of this machine, the chat reads the change sets of the chat that the call made from the shared store and puts the files on the step's card, next to the card that says who served the call (session.SetToolFiles; files with path, change set and op on the card's new files field). The step lists them as chips that open the preview pane, the chat's clients are told that its files changed, and the files reach the file list, so the answer gets its chips and the pane opens on the newest one, as for write_file. Assisted-by: Claude:claude-opus-5-5 * Keep JSON out of the answer of a trigger run The workspace skill told the model to put the widget data shape and a source marker in a trigger prompt. Some proposals then asked the run to print that JSON, and every run ended its chat with a stray line of it. The skill now says source and note live inside the data file and are never the answer, and that a prompt must not ask for JSON. The runner appends one closing instruction to the first message of every trigger run: finish with one short plain sentence. The propose_trigger and update_trigger descriptions say the prompt must stand alone and must not ask for JSON. Connector chats are talks with people and do not get the line. Assisted-by: Claude:claude-sonnet-5-5 * Read the images of a page as base64 for its frame The bridge of a page reads UTF-8 text only, so the UI could not put a png or a jpg of the page's folder into the page. A read of the page file routes with encoding=base64 now returns an image as base64 with its media type. Only the extensions of an image are read this way (png, jpg, gif, webp, avif, bmp, ico, svg); any other file is 415, so this is no way to read an arbitrary binary file. The path checks, the 1 MiB limit, the viewer's rights and, for a shared page, the consent check are the same as for a text read. Assisted-by: Claude:claude-opus-5-5 * Let a page use the files next to it A page runs from a srcdoc with no network, so fetch('news.json'), <script src="app.js">, <link rel="stylesheet" href="style.css">, <img src="logo.png"> and import './x.js' all failed without a word. Models that split a page into files, or load its data with fetch, made pages that showed nothing. Before the frame loads, the UI now reads the files that the page names with relative URLs, with the same calls as the bridge, and puts them in: classic scripts in place (deferred ones at the end of the body), style sheets as <style> with their @import and url(), images as data: URLs, and ES modules as blob: URLs behind an import map, so cycles and import() work. A page with modules gets blob: in script-src; no other part of the policy changes. Inside the frame, fetch and XMLHttpRequest read relative URLs with files.read and answer with a real Response (200, 404 for a missing file, 413, 415); any other URL rejects with "network is off in pages; put the data in a file next to the page". Every path is resolved inside the page's folder: .., %2e%2e, absolute paths, backslashes and protocol-relative URLs are refused, and the query and the fragment are dropped. The files go in only after a shared page is allowed, with the viewer's rights. The files that go in are at most 4 MiB, images 2 MiB; a script or style sheet that does not load shows a notice in the page. A change of one of these files reloads the page, as a change of index.html does. The skills, the docs and the shared pages fixture say and test the same. Assisted-by: Claude:claude-opus-5-5 * Return tool descriptions from an MCP connector test A connection test listed only the tool names. A person who adds a server cannot tell from a name such as "query" what the tool does or whether to keep the connector on. The test response now also has tool_details: each tool with the description its server gives, in the order of tools. tools is unchanged, so existing clients keep working. Assisted-by: Claude:claude-opus-5-5 * Add jq, ripgrep, sqlite3 and archive tools to the Sandbox guest The guest user cannot install packages and apt is blocked by the proxy, so any tool the model needs has to be in the image. Add jq, ripgrep, sqlite3, zip, unzip, file and tree to the final stage. A new check-tools.sh runs in the existing build step and fails the build when one of these programs, Poppler, git or curl is missing or does not start. The list is one constant, GuestTools, in the shared sandbox package. The Commands block of the prompt of an isolated chat names it. The tool description (capped at 400 characters, already full) and the turn note (cut at 400 characters) have no room for it. Tests keep the Dockerfile, the check script and the Sandbox documentation in line with the constant. Assisted-by: Claude:claude-sonnet-5-5 * Keep the helper files of a chat in its work folder A chat asked for a Word document left a 19 KB build script at the top of the user's home, next to the user's own files. Chats like it leave scripts, intermediate data, logs and drafts at the top level or in folders such as scripts/, and nothing tells them apart from what the user asked for. Each chat now has a work folder, Chats/<chat folder>/work/ in the user's home, next to its attachments. The chat takes the name of its folder at its first turn (checked against the folders and the other recent chats of the user, so two chats with one title get two names) and makes the folder only when it is used: before a bash command runs, or when a file is saved there. The prompt of every turn names the folder and says what goes in it: scripts, helper code, intermediate data, logs and drafts go there; the deliverable goes where the user said or at the top of the home; no new top-level folders for scripts; a run-once command needs no file. A sub-agent is told its parent's folder. The descriptions of bash (the Sandbox's, within its 400 characters, and the built-in one) and of write_file end with one short hint, and the built-in office, apps and workspace skills say the same in one line. The steering is soft. When write_file, move_file or a command leaves a helper file outside the work folder, the result gets one line that suggests moving it, at most once a turn of a chat. One function (workspace/workfiles) decides the kind of a file: working for anything in a chat's work folder, or for a script, a log or a temporary file by its extension or a folder named scripts, tmp or scratch; deliverable for everything else, and always for pages, the LocalAI folder, the shared folders and the files the user attached. The chat's file list (kind), the files of a command's card (files[].kind) and the card of a file tool (file_kind) carry it, so the interface only renders it. The runtime setting chats_work_folder (on by default) turns the steering off. Assisted-by: Claude:claude-opus-5-5 * Stop telling the model that the Sandbox has npm The guest image has no Node, so the network hints named a tool that does not exist. Assisted-by: Claude:claude-sonnet-5-5 * Update the Sandbox description test after dropping npm Assisted-by: Claude:claude-sonnet-5-5 * Open the links of a page after the person confirms them A page runs in a frame without allow-popups or top navigation, so a click on <a href="https://..."> did nothing, and a link to #id loaded LocalAI's own address in the frame, since a srcdoc resolves it against the parent. A news page had only dead links. The bridge of the frame now takes the clicks on links (also with a modifier, the middle button or Enter) and window.open(), and sends link.open(url) to the host. The host lets through only an http or https address without a user name or password, at most 2048 characters; javascript:, data:, file:, blob:, //host and links to other files of the page are refused with a message in the console. A bar above the page shows the site in large type, the whole address, and Open and Cancel. A host in punycode also says how it reads, since it may imitate another site. Open is a click in the app, so the new tab is not blocked and opens with noopener and noreferrer. A page asks at most 5 links at once, then one every 3 seconds. A link to #id scrolls the page. The sandbox, the policy of the frame and the consent of shared pages do not change; a shared page asks the same way. Assisted-by: Claude:claude-opus-5-5 * Let the items of a Home widget link A list widget showed each item as plain text, so a digest of articles or a list of pages could not take the person anywhere. An item now has an optional url. An http or https address without a user name opens in a new tab with noopener and noreferrer, and its tooltip names the site; a path of the app (/app/pages, /app/chats, /app/files and the other screens a person reads) opens inside LocalAI. The label is the link, with a focus ring, an underline on hover and focus, and an icon for an address outside the app. The widget-data schema checks the url, and the server adds an error in words that names the item and the rule, so a trigger run that writes a bad link learns why. The skill and the docs show the field. Assisted-by: Claude:claude-opus-5-5 * Attach any file in a chat with one button The chat had three attach buttons (image, audio, file) that all add to the same list, and the file button's dialog filtered on */*, which some file dialogs read as a type filter that greys out archives such as .zip. A chat now shows one Attach file button whose dialog has no type filter. Composers without anyFile keep the three buttons. Assisted-by: Claude:claude-opus-5-5 * Accept both Node behaviours for a broken punycode label in the link test Node 22 refuses the address, Node 24 parses it. Both mean the page never opens it unwarned. Assisted-by: Claude:claude-sonnet-5-5 * Show a chat's working files apart from what it made The Files pane, the steps and the preview treated a helper script like the document the user asked for: a 19 KB build script was listed next to the document, got a chip under the answer, and the preview opened on whichever file was newest. The interface now renders the kind the server puts on each file. The pane lists what the chat made first and folds the working files into one quiet heading under the rest, "Working files · 2", with a line that says what they are; opening it shows their rows, which preview like any file. A working file opened from a step unfolds the group. The preview that opens by itself after a turn picks the newest file that is not a working file, and the chips under the answer name only those files. A step that saved working files lists them as one line, "Also saved 2 working files", that opens to their chips, and the file of a write_file step that is a working file is a quieter chip. The files the user attached are unchanged. Assisted-by: Claude:claude-opus-5-5 * Add a search field and a segmented filter to the UI kit The Agents views each drew their own toolbar: a search box capped at 320 px on the agents list, round chip filters on the triggers history, and a bare checkbox on the tasks page. Add-ons already had the look the product wants (a search that fills the row and a segmented tray), but only as add-on classes. SearchField and Segmented are that toolbar as two small primitives with ui-* classes and token values only. Segmented buttons carry their id in data-filter, so specs can press them. The README shows how to use them. Assisted-by: Claude:claude-opus-5-5 * Draw an agent task run as a history row with its output The jobs of an agent task were table rows that opened to bare <pre> blocks: no copy, parameters printed as k=v, the traces in a raw <details>, and "by manual" in the server's words. RunItem is one run of a history list: a toggle, the state chip, what it was and what happened, when and for how long, small actions, and the details that open under it. OutputBlock draws an answer or an error in mono with a Copy button. JobRow and JobDetail use them for a job: the values it ran with as labelled chips, the answer or error, the steps, the webhook outcome and a link to the chat that holds the run. Times show as words with the full date as a tooltip. The agent task helpers gain the pure rules the page needs: the jobs of a task, the run filters (waiting and running count as in progress), the on/off filter and search of the task list, the field errors of a save, how a job started, and the API call that runs a task by name. Assisted-by: Claude:claude-opus-5-5 * Create, edit and run agent tasks in a side sheet The task form opened as bare fields above the list, with a checkbox for On and a Save button that was only disabled when a field was missing. Run with parameters was a second inline form in the same place. TaskSheet is a side sheet with three labelled groups (about, who runs it, what it does), help under each field, the parameters the prompt asks for as chips while you type, and an On switch. Save checks the required fields, says next to each one what is missing, and moves the focus to the first. RunSheet asks for each value the prompt names and shows the prompt as the task will receive it. Assisted-by: Claude:claude-opus-5-5 * Show the details of an agent task in a pane with tabs A task had no details view: its prompt, its runs and the way to run it over the API were only in the lede of the page. TaskPane is the kit's side pane with three tabs. Overview shows who runs the task, the values it asks for, the prompt and a curl call that runs it by name, with Copy. Runs lists its jobs. Settings holds the On switch, the webhook count and Delete. Run and Edit stay in the footer. The tabs move with the arrow keys, Home and End. Assisted-by: Claude:claude-opus-5-5 * Rebuild the agent tasks page from the shared list pieces The tasks page looked unstyled next to Add-ons and the agents list: two raw tables under uppercase eyebrows, a bare checkbox for All users, forms that opened above the list, three text buttons on every row, and on a phone a table that broke names one syllable per line. The page now uses the same pieces as the polished pages. The header has one line and New task. A toolbar holds the search, an All/On/Off filter and, for an administrator, an All users switch. Each task is an agent row: a tile, the name, two lines of description, who runs it and what it asks for, the last run's state and time with the run strip, an On switch, Run and a menu. A task with no parameters runs at once. The name opens the details pane; New task, Edit and Run open the sheets. The run history is grouped by day, with a state filter, a task filter and Show more. A job is deleted after an in-page question. With no tasks the page explains what one is, with an example; a failed load shows the error with Try again. On a phone the controls grow to 44 px and the filters scroll sideways. New strings, and the task strings that existed only in English, are in all nine locales; the strings the page no longer uses are gone. A new stubbed spec covers the list, filters, sheets, pane and history, and the real-server spec follows the new controls. The docs describe the page. Assisted-by: Claude:claude-opus-5-5 * Rebuild the MCP connectors page on the Add-ons patterns The page read as a form: every connector was a row of raw controls (a checkbox, a Test button, red Delete text), the add form pushed the list down, a test result vanished on reload, and the frame was narrower than the pages next to it in the sidebar. The page now uses the same blocks as Add-ons and Files: - Full page width and the shared PageHeader with a one-sentence purpose and "Add connector". - The Add-ons toolbar (search and a state filter with counts) from four connectors, and the Add-ons rows: a monogram tile, a state chip (Connected, Needs sign-in, Can't connect, Not tested, Off), one line with the tools or the reason, a meta line, the kit Switch and a menu. - Details in the shared SidePane: the switch, the last test, the tools with their descriptions and a search, the connection with the query of the address hidden, and what is saved. - Add and edit in the shared SideSheet: kind first, grouped fields with inline checks, masked secrets with show and hide, then "Save and test", the result with the tools, and Turn on. Escape and Cancel ask before they drop typed changes. - An empty state that says what a connector is for. The last test of each connector is kept in this browser with a hash of what it tested, never the address or a secret, so the page shows the last known state after a reload. Strings are in all nine locales. Assisted-by: Claude:claude-opus-5-5 * Document the MCP connectors page and its states The MCP servers section now describes the add sheet (kind, checks, masked secrets, Save and test, Turn on), the states a connector shows, where the last test is kept, the search and the filter, and the details pane. The test example shows tool_details. Assisted-by: Claude:claude-opus-5-5 * Render a preloaded page without its loading placeholder A page that was preloaded still showed the route's loading placeholder when it opened: React.lazy suspends once on its first render even when the chunk is already in memory, and React keeps a fallback on screen for at least 300 ms once it shows one. Opening a chat from Home showed a blank page for that time. The page loader now remembers the module once it has loaded and hands it to React.lazy as a thenable that resolves at once, so a preloaded page renders in the same commit. A preload of a segment now warms every page registered under it: /app/chats is the chat page, its connectors and its conversion page, and before only the last one registered was preloaded. Assisted-by: Claude:claude-opus-5-5 * Move the composer from Home into the new chat in one gesture Sending the first message on Home was a hard cut. The page stood still while the chat was created, then showed a blank placeholder, the empty "new chat" screen with its composer in the middle, an empty thread with the composer at the bottom, and only then the message, while the sidebar folded and moved the whole column. Focus was lost. Now Send fades the rest of Home and keeps the composer in place while the message goes. When the send has answered, a view transition moves the composer to the dock of the chat, lifts the text out of it into the message, moves the attached files with it and folds the sidebar in place; the model's indicator comes in after the message has landed and the empty composer has the focus. The chat page draws the message from what Home already knows (a small hand-off store: the chat record, the text and the files, with the local copy of an image for its thumbnail) and keeps the same row when the server's transcript replaces it, so nothing is drawn twice or animates again. What is sent to the server does not change. A browser without view transitions gets a FLIP slide of the composer and the message (Web Animations, transform and opacity only). Less motion gets a short cross-fade, or an instant switch. A send that fails brings Home back, shakes the composer and keeps the draft; a second Enter while one send is in flight sends nothing. The draft is cleared when Home closes, so the box never shows its placeholder on the way. Focus mode folds the sidebar in a layout effect, so the fold happens in the frame the conversation appears. One motion token is added, --dur-glide (360 ms), for a part that travels to its place on the next page. Assisted-by: Claude:claude-opus-5-5 * Draw the triggers as rows and their history like the task runs The triggers page used a table with a switch column next to the agents card list, folded its list behind Hide, filtered the history with round chips, and on a phone cut each history row at the right edge. Each trigger is now an agent row: a clock or folder tile, the name, the schedule in the server's words, the agent or model and when it runs next, the last run's state and time with the run strip, an On switch, Run now and a menu. The week sentence sits under the Run history title. The history uses the segmented filter and the same run rows and day headings as the agent tasks, with Show more after 30 runs; an open run shows its reason with Copy. On a trigger page the prompt comes before the runs, in the body face, the On switch is the kit switch, and the runs stop at 10 with Show more. Assisted-by: Claude:claude-opus-5-5 * Use the shared search and the full width on the agents list The agents list stopped its search at 320 px and used the content width, while the Triggers and Tasks views next to it use the full width, so the title and the view tabs moved when you switched views. The list now uses SearchField, which fills the toolbar row as on Add-ons, and the full page width like its sibling views. Assisted-by: Claude:claude-opus-5-5 * Polish the agent task sheets and pane after a visual review Screenshots of every state showed a few rough edges: - The name placeholder looked like a filled-in value next to the error "Give the task a name". The name field has no placeholder now, and the prompt placeholder starts with "For example:". - A task run by an agent showed the agent id until the agents were read. It says "Agent" until then. - A run opened in the details pane also opened in the history, and the two copies shared an element id. The pane keeps its own open runs and ids. - A failed run in the pane drew its meta line in red; the title carries the error colour there now. - A group title that is not a legend had a double gap under it, and the kebab menus were under 44 px on a phone. Assisted-by: Claude:claude-opus-5-5 * Cover the code files of a shared page in its consent The consent to a shared page was kept against the hash of index.html and page.json. The UI also puts the scripts, style sheets and modules next to the page into it before it runs, so the author could change app.js after a viewer allowed the page and run new code without asking again. The server now finds the same files as the UI's inliner: every script src, stylesheet link, relative CSS @import and module import of the folder, followed through the module graph. Their content hashes go into the hash of the page, and a file the page names that is missing counts too, so adding it later also asks again. Data files and images stay out, so saving data does not ask again. The files are listed as code_files. What a file names is kept per content hash in a small per-frontend cache, which only saves reads. Two more checks close the gaps around it. A frame of an older version that reads a changed file of code gets 409, so a script edited while the page is open never runs under the old consent. The UI puts no text file into a shared page that is not in code_files, so if the server and the UI ever disagree, the file fails to load instead of running outside the consent. The trust sheet names the files of code when only one of them changed. Assisted-by: Claude:claude-opus-5-5 * Let a page be a tile on Home in its compact mode Home widgets could show a list, a number or the last chats, but not one of the pages a person or the assistant built. A page can now be a tile. A page opts in with a widget block in page.json: a size (small, medium or large, the height of the tile) and an optional short title. A widget of type page in LocalAI/home.json names the page, and a folder id for a page of a shared folder. The server takes at most three page widgets and says why in words, and refuses a name that is not a page name. The tile runs the page in the same sandbox and bridge as the full view. The page reads localai.page.mode ("widget" in a tile, "page" otherwise) and localai.page.size, and <html> carries data-mode and data-size, so the page decides how its tile looks instead of being shrunk. The tile reserves the height of its size, builds its frame only once it scrolls into view, tells the page through a new visibility event when it goes out of sight or the tab is hidden, takes the frame down far off screen, and keeps at most three frames alive. It has a title bar with Open page and a refresh button, a skeleton while it loads, Retry on an error, and the link question of the full view. A page that is gone or has no widget block shows a calm line with a link to Customise Home; a shared page that the viewer has not allowed shows Allow this page, which opens the same trust sheet. Customise Home lists the pages with a widget block, with the size of their tile. The task board starter has a compact view, the skills tell the assistant how to put a page on Home, and the docs describe it. Assisted-by: Claude:claude-opus-5-5 * Give the Files upload its own size limit The Files page upload was capped by the general --upload-limit (15 MB by default), although a workspace file can be 256 MiB. Make 0 the default of --upload-limit, meaning not set. When it is not set, the general routes keep the 15 MB limit and the Files upload route skips it, so the workspace file size is its limit. When the operator sets a value, it applies to the Files upload too, lower or higher (a file never exceeds the workspace size). The max_file_size of a folder listing reports the limit in force. Assisted-by: Claude:claude-sonnet-5-5 * Refuse writes of code files from a page A page loads its own scripts, modules and styles, so a page that could overwrite app.js or style.css would run new code on its next load. The page bridge now writes data files only. The server refuses a write whose final segment has a code or document extension (js, mjs, cjs, css, html, htm, svg, xml, xhtml, wasm), is named index, or is page.json, with 403 and a clear message. The check applies to the page routes only; the Files page and the assistant's file tools keep full write access. The judgement lives in pages.IsCodeFile and is mirrored by isCodeFile in the web host, which refuses early with the error code forbidden. Both run one table of cases, covering case, double extensions, trailing dots and spaces, percent-encoding, backslashes and lookalike dots. Assisted-by: Claude:claude-sonnet-5-5 * Make the trigger tools a built-in group a chat can switch off propose_trigger, list_triggers, update_trigger and delete_trigger were added to every chat a person starts, with no switch. They are now four entries of the built-in tool registry in one group, triggers, so the Tools menu, the tools panel, the draft list and the builtins switch endpoint handle them like the web group. The group is on by default. A trigger or connector run and a sub-agent still never get them. The panel lists the group as not usable, with a reason, for those chats, for a user without the triggers feature, and not at all when the server runs no triggers. The group has no runtime setting, so the registry gets an Unlisted flag and Settings > Tools skips it. The panel now builds every grouped tool to learn whether it is usable, not only the opt-in ones. Assisted-by: Claude:claude-sonnet-5-5 * Show the Triggers group in the chat Tools menu The menu rows come from the groups the server lists, so the row needs only its label in the nine locales. A Playwright spec covers the row, its switch request, Select all and Disable all, and the disabled row with its reason. Assisted-by: Claude:claude-sonnet-5-5 * Talk about settings, not the files behind them The settings cards, the shortcuts row, Home and the change notes in a chat named the files that hold a person's settings: "Defined in LocalAI/shortcuts.json", "Layout is in LocalAI/home.json", "Saved LocalAI/persona.json". A person changes their shortcuts or their Home, not a file, and the Files page still shows the files to whoever looks. The "Stored in", "Defined in" and "Layout is in" lines are gone; the history line keeps Versions, Revert and the Invalid badge. The notes say what changed ("Saved your shortcuts.", "Created the page Task board."), the errors name the setting or the widget instead of a path, and the Source view says "The settings as text" instead of JSON and schema. A failed save now says what happened in one plain sentence and keeps the server's own words under a collapsed Details. The drafts that "Edit with the assistant" puts in the composer no longer carry the path; the built-in workspace skill already maps shortcuts and Home to their files. All nine locales are translated; only the changed keys moved, and four keys that nothing reads any more are removed. Assisted-by: Claude:claude-opus-5-5 * Drop internal names from the skills, add-ons and tasks copy Other pages used words that only make sense to someone who knows how LocalAI is built: - Skills said "a folder with a SKILL.md" and "the file workspace". It now says what a skill is for, and that skills need Files. - The add-on notes blamed "the owner", the internal name of the server that runs add-ons. They now say what happened to the action, or that the add-on service is not available; a private library names a credential "an administrator configured on this server". - The Tasks header listed an API route, and the first-run wizard spoke of a "host bash fallback". The memory line of Settings no longer mentions "the resources endpoint". All locales that have these keys are translated. The skill delete dialog, the unused skills intro and the Tasks header exist only in English and change only there. Assisted-by: Claude:claude-opus-5-5 * Fail the unit tests when UI copy names implementation details A new test reads every string of every locale and fails when one names a settings or data file (.json, .yaml, LocalAI/), an API route or HTTP method, a setting key (chats_*), an environment variable, a format word (JSON, YAML, schema) or a path shown as code. Plain words such as "schema" and "endpoint" are checked in English only, since other languages use them for other things. Operator pages may still name the model configuration, the metrics route or a startup setting that the docs name. Those keys are listed in ui-copy-allowlist.json, each with the audience and the reason. The list only shrinks: an entry whose string no longer needs it, or whose key is gone, fails too. Assisted-by: Claude:claude-opus-5-5 * Name the triggers group in the chat tools routes The routes that list and switch the built-in groups of a chat said only "a name such as web". They now name all three: web, triggers and saved_agents, with the tools of each. Assisted-by: Claude:claude-opus-5-5 * Make the built-in skills answer in plain words The skills told the model to report file names, folders and formats after a change. Add one rule near the top of localai-workspace and localai-apps: do the work with files, but say what changed, and name a file only when the user did or asks. Replace the example replies that carried paths, keep every path that says where to write, and pin the rule with a test. Assisted-by: Claude:claude-sonnet-5-5 * Apply the Go 1.26 modernizers to the latest Studio code go fix suggested slices.Contains, strings.SplitSeq and range-over-int on lines added since the first pass. ParseSearchProviders had the same hand-written membership loop, so it uses slices.Contains too. Only lines that the branch added since then are changed. Assisted-by: Claude:claude-opus-5-5 * Share the blob wait and the counting of the two workspace purges The JSON and GORM stores each had the 2-minute default for the blob grace, the same switch that reads the BlobGrace option, and the same loop that counts files, folders and bytes per revision. They now call blob.Grace and PurgeResult.Add, both with specs. The SQL and the in-memory walk that pick the nodes stay in each store. Assisted-by: Claude:claude-opus-5-5 * Read the upper layer of a sandbox in one place for both stores The JSON and GORM workspace stores had the same readUpper, place, joinRel, depth and sortedKeys, with the same upperSet types. They move to workspace/internal/upperset, which has its own specs. The commit algorithm that applies the set to the nodes stays in each store; it needs a storage interface first. Assisted-by: Claude:claude-opus-5-5 * Let chat attachments use the 1 GiB message limit The docs promise 1 GiB for a message with attachments, but the general body limit (15 MB by default) ran first, because only the Files upload skipped it. Skip it also for a multipart POST to the messages route when the operator set no --upload-limit. A JSON message keeps the general limit, and an explicit --upload-limit still wins. A message that goes over the 1 GiB limit now gets 413, not 400. Assisted-by: Claude:claude-opus-5-5 * Name the read_file tool in the skills index header A weaker model read a built-in skill with the wrong path and then tried the Sandbox, which cannot see skill folders. The header now says which tool reads a skill and that the shown path is the one to use, before it mentions where the user's own skills go. Assisted-by: Claude:claude-sonnet-5-5 * Tell the model that skill folders are not in the Sandbox The Commands block of an isolated chat now says that skills are read with read_file. The Sandbox tool description and the per-turn note are at their size caps, so the sentence goes in the system prompt. Assisted-by: Claude:claude-sonnet-5-5 * Point read_document at read_file for text files read_document reads PDF, Word, Excel, PowerPoint and EPUB files. For a Markdown or text file it said only that the file is not a document it can read. The error now names read_file. Assisted-by: Claude:claude-sonnet-5-5 * Tidy the text index stores and the indexer pool The two stores wrote the default limits 20 and 100 as numbers in five places; they are now named constants, so the stores cannot drift. The JSON store checks a missing file with errors.Is. The indexer starts its readers with wg.Go and keeps the first error inline, which drops the cmpErr helper. Assisted-by: Claude:claude-opus-5-5 * Refuse the server's public address in the Sandbox proxy web_fetch refuses the addresses of the server: its interfaces, loopback and the host of the external base URL. The Sandbox proxy runs in the add-on process and listed only the interfaces of its own machine. Under "All public sites", an always-allowed host or an Allow card, a command could reach the LocalAI API through the public name or address of the server, a NAT address or a proxy in front of it. The host now keeps one list for both. The application builds it once, gives it to web_fetch, and sends it as CIDR strings in the new self field of every egress.check allow. The proxy keeps the last list it got and refuses it on each dial, besides its own interfaces, also for an allow it cached. safehttp gets the codec and a join of deny lists, so both sides read the list the same way. Assisted-by: Claude:claude-opus-5-5 * Check the TLS server name on every port of the Sandbox proxy The proxy checked the ClientHello only on CONNECT to port 443. An always-allowed host pinned to another port, or any port added to chats_egress_ports, was a raw tunnel: a command could open TLS to another tenant of the same front through it. The docs said the server name must still match. On other ports the proxy still waits for nothing, since the protocol may be anything and the server may speak first. But when the first bytes from the guest are a TLS handshake record, it reads the whole ClientHello and passes it on only when the server name is the CONNECT host; otherwise it closes the tunnel and audits the reason sni. Assisted-by: Claude:claude-opus-5-5 * Say what sent each notification The inbox only knew whether a notification was about a run or needed attention, so a trigger that was switched off and a connector that was switched off looked the same. Each notification now carries a source: trigger, connector or system (the conversion of old data). The list endpoint returns it, and items stored before have none. Assisted-by: Claude:claude-opus-5-5 * Say where a skill is when a read of it fails When read_file, read_document or list_files fail with not found for a path that names a skill of the chat (a wrong skills folder, or just <name>/SKILL.md), the error now gives the skill's real path and says to read it with read_file. The skills come from the chat's own resolver, so every layer, add-on and per-user switch is respected, and nothing is looked up unless a read fails. Assisted-by: Claude:claude-sonnet-5-5 * Keep the last valid Sandbox network lists when a value is bad The settings endpoint refuses a bad chats_egress_allowed_hosts with the reason, and the runtime settings file keeps the value in force. A bad value that reached the configuration another way was dropped with no log line: the always-allowed hosts and the extra registries became empty, and a bad port list fell back to 80 and 443, which can widen a narrower list. Each of the three lists now keeps the last value that parsed, logs a bad value once, and uses the default only when the process never saw a valid one. The parse runs again only when the value changes, not on every egress.check. Assisted-by: Claude:claude-opus-5-5 * Use the keyless chain of the built-in search in the Web add-on The add-on replaces the built-in web_search, but its keyless provider was DuckDuckGo alone. DuckDuckGo often answers data-center addresses with a bot check, so installing the add-on without a key made search worse than the built-in, which falls back to the Brave Search and Bing result pages. The keyless entry of the add-on's order (saved name duckduckgo) is now the built-in chain from pkg/web/search: DuckDuckGo, then the Brave Search page, then the Bing page, with the same cool-down. The status tool reports the providers in their cool-down, and the Settings tab shows them with the reason. Test the providers gives one row for each page of the chain. Assisted-by: Claude:claude-opus-5-5 * Fix the GitHub hosts in the Q1 row of the egress design The Q1 row still listed raw.githubusercontent.com on the registries list. The owner decision below it and registries.json exclude it, since it serves any user content, and add release-assets.githubusercontent.com. The row now says the same. Assisted-by: Claude:claude-opus-5-5 * Keep one private-address rule for the network cards A card asked whether its host is a private IP literal by calling egress.Admits with a made-up ceiling, which re-derived the rule that egress already has. egress.PrivateLiteral is now exported (with a spec) and the card uses it. The two String methods of an allowed host write host:port with net.JoinHostPort, which brackets an IPv6 literal the same way the hand-written code did. Assisted-by: Claude:claude-opus-5-5 * Read a unique constraint violation in one place The jobs and triggers stores had the same isUniqueViolation, and the saved agents store had the same test inline before it looked for the name index. They now call dberr.IsUniqueViolation, which has specs for the SQLite and PostgreSQL error texts. The workspace store keeps its own stricter match. Assisted-by: Claude:claude-opus-5-5 * Ask before a Home widget link opens another site A list widget turned an http or https item into a plain link to a new tab. Its data file is written by trigger runs and the assistant, which read web text, so a planted link opened with one click, and only a tooltip showed the host. A link of a page asks first. An external item link now opens the confirmation bar of page links in the widget: the site, the full address, the punycode warning, and Open or Cancel. A link inside the app still navigates directly. The bar takes a label, so the widget says that it wants to open the link. Assisted-by: Claude:claude-opus-5-5 * Prune the audit of the commands run on this machine The audit of the built-in bash had no retention: the database table grew without bound in a cluster, with commands of up to 4 KiB each. The Sandbox network audit had chats_egress_audit_days and a daily prune under an advisory lock. Both audits record what the commands of the chats did, so the host command audit now uses the same retention and is pruned in the same daily pass: one frontend of a cluster, under the same lock; a standalone process alone. The GORM and file stores get Prune. No new setting. Assisted-by: Claude:claude-opus-5-5 * Save the Studio JSON stores with atomicfile.WriteJSON Seven stores (agents, connectors, jobs, MCP connections, notifications, push subscriptions and triggers) encoded their state and then called atomicfile.Write with the same four lines. atomicfile.WriteJSON does both and has specs. The load side stays in each store, since the error messages differ. Assisted-by: Claude:claude-opus-5-5 * Build the result page walkers of the keyless search on one walk hasChallenge and firstByClass each had their own recursive walk that stopped at the first match. They are now predicates passed to firstElement, which already did that walk. The order of the walk and the element tested first are the same. Assisted-by: Claude:claude-opus-5-5 * Expire network cards that a dead frontend left pending Pending network cards expired only on the turn-end path of the process that ran the turn. When that frontend died, its cards stayed pending in the chat record, counted toward the cap of three, and gave the next turn's connections cards-full blocks. The UI showed them as waiting. A card now names the turn that asked it. One rule, shared with the command card in a later change, says whether a card still has a waiter: its turn holds the chat and its wait has not run out. A card that fails it reads as expired, takes no place under the cap, and is expired in the record when the next card is asked, when the next turn starts, or when the reaper ends the dead turn. Assisted-by: Claude:claude-opus-5-5 * Reuse ChooseProvider for the expected add-on of a tool shadowers.expected repeated ChooseProvider line for line: the pinned add-on when it is installed, else the oldest install. It now calls it. commandFiles kept an order field and sorted the nodes of a map by it; it now appends each node to a list when it first sees it, which gives the same order without the sort. Assisted-by: Claude:claude-opus-5-5 * Do not print a nil error when a key or entry is missing Three checks joined "the read failed" and "the read found nothing" in one branch, so the second case wrote "<nil>" into the message: the VAPID keys of push notifications, the sealing key of connectors and a rendered catalog entry. Each case now has its own message. The message of a failed read does not change. Assisted-by: Claude:claude-opus-5-5 * Check the close errors and rename the tool variables golangci-lint failed on the new studio code. Five deferred Close calls and one Fprintf into a hash dropped their error, and two locals named t made forbidigo read `t.Run` as a testing call. Discard the errors explicitly and name the locals tool. Assisted-by: Claude:claude-opus-5-5 * Make the file text index forget purged files and say its limits Three gaps of the text index of documents: - A purge dropped the index entries of what it erased in a goroutine that only logged its errors, and the purge had already erased the journal rows. A crash or a store error left the text in the database for good. Each pass of the indexer now also checks a batch of entries (200 every 30 s, round all of them) and deletes those whose node is gone, so what a failed Forget leaves is removed. - A document search stopped at 20000 files or 64 MiB of text without a word, and the "still being indexed" count stopped at 50. The search now sets Limited (search_files says "results limited", the REST answer has "limited", the Files page shows a note), and the count is the number of pending entries below the path. - read_file extracted a long document again for every page, up to 50 MiB and 60 s each time. The first read keeps the whole text in the index store by content hash (a new documents table, or files in the JSON store), and the next pages come from it. The text goes away with the last entry of that content, on delete and on purge. Search reads the text without its folded copy, which only the database needs. Assisted-by: Claude:claude-opus-5-5 * Stop the PDF reader when the extraction gives up The PDF and EPUB reader runs in a goroutine, and the call returns after 60 seconds when it is slow. The goroutine went on reading every page, since its context ended only with the caller's. It now gets a context that ends when the call returns, so it stops at the next page. What the call returns does not change. Assisted-by: Claude:claude-opus-5-5 * Compare the code files of a shared page as the store does The guard that refuses new code to a frame of an older version compared the name it reads with the names of the page's code exactly. The workspace folds case, so an old frame that read Index.html or JS/App.js got the new code with no 409. Compare names in the store's form (workspace.FoldName), as the write check already does, and test the case variants. Assisted-by: Claude:claude-opus-5-5 * Expire the command card of a turn that is over The card of the built-in bash was cleared only by a defer in the call that asked. When that frontend died, the card stayed in the chat record. The chat hid it only while no turn ran, so the next turn showed it again, and an answer to it reached no one but still got 204. The card now names its turn and when it was asked, and follows the same rule as the network cards: it waits only while that turn holds the chat and its five minutes have not run out. GET /api/chats/{id} shows only such a card. The next turn and the reaper clear a card left behind, and the waiting call clears only its own card, so it never removes a newer one. An answer to a card with nobody waiting is 410 consent_expired, and the chat shows the card as not answered, with no buttons. The store gets UpdateHostConsent, an atomic change of the card. Assisted-by: Claude:claude-opus-5-5 * Answer a trigger change once across frontends ApplyChange, UndoChange and DismissChange checked the outcome of the change record under a mutex of one process, then wrote the store. Two frontends could both see a pending change and both apply it, or both undo it. Add Store.SetOutcome, a conditional update of the outcome (only from the outcome it names), to the GORM and JSON stores and to the store suite. The three answers now claim the record with it before they write, give the outcome back when the change itself fails, and answer ErrProposalAnswered (409) when another request won. A test runs the two answers on two frontends at the moment between the read and the write, for an edit, its undo, and a delete. Assisted-by: Claude:claude-opus-5-5 * Add the shadowers spec to the tables-tools inventory The tool adapter gained a spec that lists the add-ons that replace a built-in in any state. The tables-tools suite runs every spec of that package, so verify.sh found one case more than the inventory holds. Assisted-by: Claude:claude-opus-5-5 * Offer Clear filters when the backend Catalog shows nothing The Installed view had a Clear filters button in its no-results state; the Catalog view did not, although the spec asks for it in both. Move the button to one place and show it in the Catalog when a search, a type filter or a library filter leaves no row. Add Playwright specs for both views, which the earlier Installed fix lacked. Assisted-by: Claude:claude-opus-5-5 * Give the chat settings without an assistant tool their reason chats_max_iterations, chats_work_folder, chats_index_files and chats_index_max_text_bytes had no MCP tool and no entry in adminRoutesWithoutTool, and the older chats_saved_agents, chats_suggestions, chats_vision_model and chats_transcription_model were unlisted too. Add each with the reason it stays a person's choice in Settings: they apply to every user's chats, and the step limit is one a model must not raise for itself. Assisted-by: Claude:claude-opus-5-5 * Remove the stray allowed-hosts keys from the chats inbox locales allowedHosts, allowedHostsOff and allowedHostsNote sat twice in eight locales: under egress, where English has them and the UI reads them, and again under inbox, where nothing reads them. In de and ko the two copies had drifted apart. Delete the inbox copies and keep the egress ones. Assisted-by: Claude:claude-opus-5-5 * Bring three design specs in line with what shipped - web tools and shadowing: the status still said phase 2 waits; both phases shipped. Note that the Web add-on's keyless entry is now the built-in keyless chain. - Home front door: Home now creates the chat and sends the first message itself, with a short-lived hand-off to the chat page, and its composer takes attachments. Both were described the old way or as out of scope. - chat footer: the meter bar is 96 px with the tokens left, and a phone keeps the bar and the tokens left, not the percentage only. Assisted-by: Claude:claude-opus-5-5 * Store the command card's answer and read it back every two seconds The answer to the command card of the built-in bash went only as a bus message. When the message was lost, the call waited the full five minutes and ended as not answered, although the person had clicked. The network card stores its answer first and re-reads the card every two seconds. The command card now does the same, through one shared pair of helpers: listenConsent subscribes before the card is stored, and waitConsent wakes on the bus message or every two seconds and reads the record, which holds the truth. The network card uses the same helpers. HostExecDecide writes the decision to the card atomically before it sends the message: the same answer twice is 204, the other answer 409 already_answered, and an answered card is no longer shown as waiting. Assisted-by: Claude:claude-opus-5-5 * Load skills by name with a load_skill tool A weaker model failed to read a skill at the path the index showed. The chat now has a built-in load_skill tool that takes a skill's name and returns the instructions of its SKILL.md, without the frontmatter. The tool reads the same resolved skill set as the index: the nearest skill of each name, the chat's switches, the add-ons the chat can use now and the grants of its agent. It reads through the user's rights. When the skill's folder holds other files, the answer starts with the base directory and names up to 20 of them for read_file; a chat without read_file gets the instructions only. A skill that needs commands, in a chat that cannot run them, is refused with the reason. An unknown name is an error that gives near names first and then every name the chat has. Every chat with at least one skill gets the tool, and a sub-agent gets it with its parent's tools. It has no switch and Settings > Tools does not list it. The index of skills now lists names and descriptions only and tells the model to call load_skill. The Sandbox sentence and the error for a wrong skill path now point at load_skill too. Assisted-by: Claude:claude-opus-5-5 * Tell the built-in skills to load other skills by name localai-apps told the model to read the localai-workspace skill, and writing-skills said a description is what the agent sees before it opens the file. Both now speak of loading a skill with load_skill. localai-workspace keeps the path of its schemas folder and says to read it with read_file, since those files are not a skill. A test checks that no built-in skill sends the model to another skill by "read" or by the path of its SKILL.md. Assisted-by: Claude:claude-opus-5-5 * Document load_skill and the skills index The chats page now says that the prompt lists skills by name and description, that load_skill returns a skill's instructions and names the files of its folder, what it refuses and who gets it. It no longer says the model reads a skill by its path. The add-ons page says how a chat loads an add-on's skill, and the saved agents page says that the skills grant limits load_skill too. Assisted-by: Claude:claude-opus-5-5 * Add the proposal action spec to the tables-acceptance inventory The spec that an editor cannot confirm a schema change as a write ran in the suite but was never in the inventory, so the evidence manifest check failed. CI did not reach that check before, because the burst spec failed first. Assisted-by: Claude:claude-opus-5-5 * Give a chat a goal that it works toward by itself A chat can now hold one goal. While it is active, every turn's prompt names it, the model gets goal_done and goal_blocked, and when a turn ends by itself the frontend that ran it starts the next one with "Continue toward the goal". The person's waiting messages still go first, and a turn they start does not count as an automatic one. The goal pauses when the person presses Stop, when a turn fails or is cut short, and when the automatic turns or the model steps reach their limits. Resume gives it fresh budgets. A sub-agent, and a chat that a trigger, a connector, a job or a page started, has no goal and no goal tools. The goal lives with the chat record in the store (UpdateGoal on the JSON and GORM stores, with contract tests and a column migration), so every frontend reads the same one. The next turn goes through the same turn claim as any other, so only one frontend continues it; a reaped turn, or a JSON store that reloads after a crash, leaves the goal paused instead of continued twice. The transcript keeps goal lines (set, replaced, continue, paused, resumed, done, blocked, cleared) as their own role, so an automatic message never reads as typed by the person. Assisted-by: Claude:claude-opus-5-5 * Add a store for the loops of the chats A loop repeats a prompt in one chat, at an interval or when the model asks for the next run. The new loops package keeps the loops, the daily budget of runs of each owner, and which chats confirmed a loop once. JSON and GORM stores share one contract suite. Add checks the limits per chat and per owner in the same step as the insert, and TakeRun never hands out the last run of a day twice, also across frontends. A user message that a loop sent now carries a loop mark (loop ID, run number, self-paced or not). Both chat stores keep it. The model reads a short line before the prompt that says the message repeats by itself, and, for a self-paced loop, how to ask for the next run. Assisted-by: Claude:claude-opus-5-5 * Add the file preview slot specs to the host-slots inventory The file preview slot came with two Go specs in the host-slots package. verify.sh found them in the run but not in the inventory. Assisted-by: Claude:claude-opus-5-5 * Check the Settings page and the add-ons with the UI copy guard The guard read the locale files only. The Settings page keeps its English labels and descriptions in settingsSchema.js, and several add-ons carry their own English bundle, so neither was checked. The guard now also reads the label, description and note of every setting and the label and hint of every group, and the English bundles of the add-ons (an en.js module or an English JSON file). The allow-list takes settings:<key>.<what> and addon:<id>:<key> entries, each with its reason; the operator ones (the SearXNG JSON format, the fine-tuning file types, the API route of a voice, the schema of the Databases add-on) are listed. What it found in the voice library is fixed in plain words: "model YAML is not modified" and "Existing model YAML is not changed" now say that the models' settings are not changed, in the voice add-on, the Studio add-on and the host locales that have the strings (en, fr, pt-BR). The two host allow-list entries for them go. Assisted-by: Claude:claude-opus-5-5 * Add the update digest specs to the updates inventory Two checker specs pin that an update names its candidate digest and that a stale cached notice stays readable. The updates suite runs the whole checker package, so the inventory needs both. Assisted-by: Claude:claude-opus-5-5 * Draw the Add-ons search and state filter with the kit pieces The Add-ons page had its own copy of the search box and the segmented filter (ad-search, ad-seg), the same shapes and styles as the kit's SearchField and Segmented that the Agents pages use. Use the kit pieces and drop the copy; the narrow-screen rules now target the kit classes inside the toolbar. The search keeps its test id. Assisted-by: Claude:claude-opus-5-5 * Add the limits of a chat goal to the runtime settings chats_goal_max_turns (1 to 200, default 20) is the most turns a chat with a goal starts by itself, and chats_goal_max_steps (1 to 10000, default 400) the most model steps of all the turns of one goal, before the goal pauses. Both are read when a turn ends, so they apply without a restart. They sit in Settings > Tools next to the step limit of one turn, and the settings page documents them. Assisted-by: Claude:claude-opus-5-5 * Add the goal routes of a chat to the API GET, PUT and DELETE /api/chats/{id}/goal read, set and clear a chat's goal, and POST /api/chats/{id}/goal/resume resumes a paused or blocked one. They belong to the chat's owner like the other chat routes and take the chats feature. A chat view carries the goal with the limits that apply now, the transcript gives goal lines with goal_event and goal_turn, and the event stream has a goal event so that other tabs read the chat again. The routes are in the route feature registry, the swagger file and the chats entry of /api/instructions. They manage the person's own chat, not the server, so they have no MCP tool, as the other chat routes. The chats page documents goals, their states and limits, and how several frontends share them. Assisted-by: Claude:claude-opus-5-5 * Redesign the notification inbox The inbox panel was a plain list: every row had the same weight, a failed run looked like a finished one, the panel ran off the screen on a phone, and a failed load showed nothing. The panel now has a header with the unread count, a menu for Clear read and Clear all, Mark all read, and an All / Unread / Failed filter. Items are grouped under Today, Yesterday and Earlier. Each item shows what sent it (trigger, connector or LocalAI) with a tinted icon, the title, two lines of text or the reason of a failure, and when it happened, with the exact time on hover. A click marks it read and opens its chat; a connector that was switched off opens the connectors page. Mark read and Delete show on hover or focus, and stay in view on a touch screen. The panel has a loading skeleton, an error with Try again, an empty state that says what shows up here, and a caught-up state. A new notification slides in, is highlighted for a few seconds and is announced once. The badge stops at 99+. Focus moves into the panel when it opens; the arrow keys, Home, End and Delete work on the items, and Esc returns focus to the button. On a phone the panel is a sheet over the bottom of the screen with 44 px targets. Assisted-by: Claude:claude-opus-5-5 * Explain the reads of the text index and the provider requests to gosec gosec flagged three new lines. The text index reads files from the directory the host gave it, with names it listed there. A search provider sends its request to the endpoint fixed at construction, never to a URL from a tool call. Mark the lines with the reason. Assisted-by: Claude:claude-opus-5-5 * Keep the add-on details open when Escape closes a dialog Escape in the restore dialog of the recovery panel also closed the details pane around it. The pane handles Escape in React, and the dialog listens on the document, so the pane ran first and saw no preventDefault. The pane now registers as a modal layer and closes only when it is the top layer, as the side panes do. Assisted-by: Claude:claude-opus-5-5 * Add skill and agent messages, compact and clear to the chats API A message can now carry a skill of the chat or one of the user's saved agents. The server checks the name against the chat's resolved skill set (switches, the agent's grants, shadowing, add-ons) or the saved agents the chat may call, and refuses an unknown one with a 400 that lists what the chat has. A skill message gives the model the text of the skill's SKILL.md before the user's words, for that turn only; an agent message asks the model to start the agent with spawn_agent. The transcript keeps the words and the command (Entry.Invoked), not the instructions. A regenerate or an edit runs the command again. Such a message does not wait for a running turn: it answers 409. POST /api/chats/{id}/compact runs the summary the loop makes when the context fills up, through a new optional session.Compactor that the loop harness implements with its own history compaction. The compacted state is saved for the next turn and a "compacted" line with the sizes goes in the transcript. POST /api/chats/{id}/clear writes a "cleared" line and drops the saved state. The model then sees nothing above the line: the rebuilt state and the transcript summary start after the last such line, so a fork from above it keeps what the model saw there. Export, import and the Markdown export carry both lines. Both calls take the turn claim, so no frontend starts a turn on the chat meanwhile; messages sent during them start a turn after. No MCP tool: they act on one chat of the caller and are not admin calls. Assisted-by: Claude:claude-opus-5-5 * Give each network card spec its own settings A check that a card spec starts can still wait on its card when the next spec begins. It read the settings through a closure over a variable that the next BeforeEach replaced, which the race detector reported. Each service now reads the settings holder it was built with, under a lock, and the specs change the settings through it. Assisted-by: Claude:claude-opus-5-5 * Show the effective audit retention in the settings snapshot The settings snapshot returned 0 for chats_egress_audit_days while the setting was unset. Any page that saves the whole snapshot back, such as the Traces settings strip, then posted 0 and the server rejected it with "Invalid chats_egress_audit_days", so saving tracing settings failed on a fresh install. Report the retention that applies (30 days by default), the same way chats_max_iterations does. Assisted-by: Claude:claude-sonnet-5-5 * Run the Sandbox smoke test in a short scratch folder The smoke test step was reached for the first time and both isolated backends refused to start: their state folder under the checkout's cache made the socket paths 126 and 132 bytes, over the 107 a unix socket allows. Point SANDBOX_SMOKE_WORK at /tmp/sandbox-smoke, which leaves room for the sockets. Assisted-by: Claude:claude-opus-5-5 * Use paths from the home folder in the Sandbox command specs bash now starts in the user's home folder, so the end-to-end specs that read and wrote home/<name> found nothing. Use names relative to the home folder, as the Sandbox specs already do. Assisted-by: Claude:claude-opus-5-5 * Show the goal of a chat on the chat page A bar above the composer shows the goal with its state: Pause (which stops the running turn and so pauses the goal) while it works, Resume when it paused or needs the person, and Clear. A second line says why it paused, what the model needs, or what it did. The status line shows "Goal · turn 3 of 20" while the goal is active, and the Continue button of a paused turn stays away then, since the next turn starts by itself. The transcript shows the goal lines as quiet rows: set, replaced, continuing (with its turn), paused with the reason in plain words, resumed, reached with the summary, blocked and cleared. A goal event, also from another tab, reads the chat again, and a hidden live region says each change to a screen reader once. The strings are in all nine locales. Unit tests cover the helpers and the thread rows; a Playwright spec covers the bar, the progress, the lines, Pause, Resume, Clear and the event. Assisted-by: Claude:claude-opus-5-5 * Run the loops of the chats The loop host fires the loops that are due. A run sends the prompt to the chat as a new message only when the chat is idle: a busy chat skips the run, and an interval loop moves to its next slot instead of piling missed runs up. A self-paced loop gets two tools while it is active: schedule_next (clamped to the minimum interval and one day) and loop_done. A run in which the model calls neither ends the loop with a note. A loop ends after its run limit, pauses after three failed runs in a row, and waits for the next UTC day when the owner's daily budget of runs is used up. One frontend fires the loops, the holder of a new advisory lock. A run is claimed in the store first, and the turn claim keeps one turn per chat, so a short overlap of two leaders starts no run twice. A new leader finishes the runs the old one watched. Deleting a chat removes its loops; a due loop whose chat is gone removes itself. Five runtime settings bound the loops: chats_loop_min_interval, chats_loop_max_runs, chats_loop_max_per_chat, chats_loop_max_per_user and chats_loop_daily_runs. Assisted-by: Claude:claude-opus-5-5 * Add the loops API of the chats GET, POST and DELETE /api/chats/{id}/loops list, start and stop the loops of a chat; PUT and DELETE /api/chats/{id}/loops/{loopId} pause, resume, confirm or stop one. GET /api/chats/{id} carries the loops, the transcript marks the user lines a loop sent with loop {id, run}, and a loops event tells every open tab that the list changed. The routes are gated by the chats feature and documented in the swagger spec and the API instructions. They act on the caller's own chats only, so they get no assistant tool. The chats page gets a loop section, with when to use a loop and when a trigger, and the cluster behaviour; the runtime settings page lists the five loop settings. Assisted-by: Claude:claude-opus-5-5 * Show the loops of a chat above the message box Each loop of the open chat shows as one line over the message box, such as "Loop · every 5 min · run 4 · next at 09:20", with Pause or Resume and Stop. A loop that paused itself or ended says why; an ended one has Dismiss. The first loop of a chat shows a card instead: "Run this every 5 minutes until you stop it?" with Start and Cancel. The server keeps the pending loop, so a reload shows the card again. A message that a loop sent carries a small "Loop run 4" label. The page reads the loops again on the loops event, so every open tab follows. The answer to /loop list is ready for the slash command. The strings are in all nine locales. Assisted-by: Claude:claude-opus-5-5 * Build the chat slash menu on one command registry The slash commands of Home and of a chat now come from one registry, src/utils/slashCommands.js. A command has a kind (navigate, insert or server), the surfaces that offer it, a needs check and, optionally, an argument provider: args(query, ctx) returns the items the menu lists when the box holds "/cmd <query>". Every existing command and test id stays; the registry adds: - one command per skill of the chat or the draft (/review), marked built-in, add-on, shared or yours, and /skill <name>, which also reaches a skill named like a command; - /agent <name> <task>, which lists the saved agents, or offers to turn them on for the chat when they are off; - /model and /models with a model name, which switch at once; - /compact and /clear, which show their progress in the status line and leave a marker in the thread; the context meter drops; - /attach and /attach clear, /help (a card that lists the commands) and /about (a card with the model, its state, the tools, skills, where commands run and the context size). Tab completes a command or an argument, Enter runs it, and the first row whose name starts with what was typed is selected first. A message typed as a skill or agent command is sent with its marker, not as text; the transcript shows the words and a chip. Home offers the same set without the commands that need a thread. The words are in all nine locales. The registry has unit tests, and a Playwright spec covers the menu, the completion, the marker, the chip, compact, clear and the cards with the routes stubbed. The docs page gains a slash commands section. Assisted-by: Claude:claude-opus-5-5 * Add the /goal command to the chat /goal shows a card with the goal, its state, the automatic turns and steps it used out of the limits, and when it was set. /goal <text> sets a new goal at once, replacing the one the chat had without a question, and the goal bar says what it replaced. /goal clear and /goal resume act on the goal; typing "/goal " lists resume and clear for the goal the chat has. None of these is sent as a message. In a new chat, /goal <text> creates the chat first. The command sits in the Context group of the shared command registry, on the chat surface only, and not in a chat that a trigger or another run started. A skill named goal is reached through /skill. Unit tests cover the parser and the menu items; the Playwright spec covers the card, setting, replacing, resuming and clearing from the box. Assisted-by: Claude:claude-opus-5-5 * Add /loop to the chat /loop [interval] <prompt> starts a loop in the chat; /loop list shows the loops of the chat with the code that /loop stop takes, and /loop stop [id] stops one or all. The command is in the slash registry: after "/loop " the menu offers list and stop, and after "stop " the loops of the chat. A loop prompt that invokes a skill (/review the diff, or /skill review the diff) runs as the person's skill command would: the model gets the skill's text, and the transcript shows the words, the skill's chip and the loop label. The server checks the skill when the loop starts and at each run; any other command is refused with the reason. Assisted-by: Claude:claude-opus-5-5 * List the loop settings on the Settings page The five limits of the loops of the chats show under Tools with the other chat settings: the shortest interval, the runs per loop, the loops per chat and per user, and the loop runs per day. All apply at once. Assisted-by: Claude:claude-opus-5-5 * Clear a sent Home draft when the page goes away too Home clears the draft of a sent message when it closes, so the text stays visible while the composer moves into the new chat. While the chat page still loads, a reload or a closed tab could come first, and the sent text came back on Home. Leaving the page now clears it too. Assisted-by: Claude:claude-opus-5-5 * Check the close error of a previewed file golangci-lint (errcheck) flagged the deferred Close of the file the preview endpoint reads. Discard its error explicitly, as the other studio file handlers do. Assisted-by: Claude:claude-opus-5-5 * Explain the document read of the text index to gosec gosec reported the read of a stored document text as a file inclusion through a variable. The path is the configured index directory joined with a hash that documentPath refuses when it holds a path separator or a dot, so it cannot leave the documents folder. Assisted-by: Claude:claude-opus-5-5 * Build the darwin/arm64 add-on packages on macos-15 GitHub retires the macos-14 runner image on 2026-11-02 and runs scheduled brownouts until then: every darwin/arm64 package job failed at once with the brownout notice. macos-15 is the Apple Silicon image that replaces it. Assisted-by: Claude:claude-opus-5-5 * Expect the chat's own load_skill tool in the agent prompt spec The spec that keeps the agent engine's text out of the prompt also required that no load_skill tool or mention reach the model. Chats now offer their own load_skill tool and the skills index tells the model to call it, so the spec failed. Keep the checks on the engine's own skill and command text, and check that the chat's tool is offered. Assisted-by: Claude:claude-opus-5-5 * Read the user's scripted tool call in the mock model The chat prompt now names the chat's work folder, and the folder name comes from the chat title, which holds a cut copy of the user message. The mock model read the first ADDON_TOOL_CALL marker, the cut one in the system prompt, and called the tool with broken arguments over and over. Read the last marker, the one of the user message. Assisted-by: Claude:claude-opus-5-5 * Keep the phone targets of the inbox at 44 pixels in every build On a phone the inbox close button and the item actions must be 44 pixels wide. Their rule had the same weight as the kit's .dk-btn min-width, so the order of the stylesheets decided. The single-chunk coverage build puts the kit later and the close button shrank to 28 pixels. Scope both rules to their container so they win in any order. Assisted-by: Claude:claude-opus-5-5 * Let the about-you spec accept the sidebar's own seen date When the user has pages, the sidebar writes the date of its New label under seen.new in user.json. In the whole suite other specs make pages first, so the saved record held that date as well and the exact comparison failed. Check that the uploaded seen record is kept. Assisted-by: Claude:claude-opus-5-5 * Give the typed-while-working chat spec time for two turns The spec waits up to 30 seconds for each step, but two turns of the mock model with its tool steps take about 28 seconds on their own, so the default 30 second test limit cut it under load. Set its limit to 90 seconds. Assisted-by: Claude:claude-opus-5-5 * Scope the foundation spec's state query to the sheet header The details pane now repeats the observed state under Health, so a role query for "State: Off" inside the add-on sheet matched two elements and failed in strict mode. The administrator-manages spec, and the two specs after it, did not run. Query the chip in the header (.ad-pane__meta), which is the state the administrator sees next to the add-on's name, through one helper used for all three assertions. Assisted-by: Claude:claude-sonnet-5-5 * Wait for the page to read each apps answer in the registry polling spec The spec moved the page clock as soon as the stand-in saw a request. The registry sets its next timer only after it reads the answer, so on a slow runner the clock could pass the due time of a timer that did not exist yet, and the next read never came. With the CPU throttled ten times, the old spec failed in 3 of 3 runs. The spec now counts the answers the page has read and lets queued work finish before it moves the clock. It pauses the clock before it leaves the page, and it tries the pause again when a slow machine passed the target time. The assertions do not change: one read at once, one read per interval and no faster, and no read after the disable. Assisted-by: Claude:claude-opus-5-5 * Close the egress test listener only after it accepts The dial helper of the egress proxy specs closed its listener when it returned, while the goroutine that accepts the connection could still be waiting to run. On a loaded runner the listener sometimes closed first. The kernel then reset the connection that was still in the accept queue, the spec's write consumed the reset error, which the spec ignores, and the next read saw EOF. The spec failed with "unexpected EOF" before the proxy got the request. Wait for Accept to return before the listener closes, then start the proxy handler on the accepted connection. Assisted-by: Claude:claude-opus-5-5 * Run host commands in a shell that reads no startup file The host backend promises a fixed environment, but bash -lc reads /etc/profile and the user's dotfiles. On macOS path_helper rewrote PATH there, so the command did not get the fixed PATH. Turn the login shell into 'bash --noprofile --norc -c' in the host backend only, and make the test assert the exact fixed PATH. Assisted-by: Claude:claude-sonnet-5-5 * Answer the command card in the platform turns and 400 a bad answer The tool-list spec of the platform suite timed out: without accounts the one person is an administrator, so the chat has the built-in bash in ask mode. The tool list is sorted and bash is first, and the mock model calls the first tool, so every turn waited five minutes for a yes on the command card. The turn helper now refuses each card it sees, as a person who clicks deny, and the turn goes on to its step limit. bash stays in the checked list. An administrator who posts a consent with no card id or an unknown decision got 403 "only an administrator can allow commands". That is a bad request, not a refusal: it now answers 400, as the swagger already says, and the route check for administrators passes. Assisted-by: Claude:claude-opus-5-5 * Let a chat switch off its shell, file, media, collection and sub-agent tools The Tools menu of a chat switched only three groups of built-in tools: web, saved_agents and triggers. A person could not keep a chat away from the shell, the files, the helper models, the collections or the sub-agents. Add five groups to the registry of built-in tools, all on until the chat switches them off: - shell: bash, whoever serves it. The Sandbox add-on's bash and the built-in bash on this machine both go, and the add-on list reports commands switched_off. The row stays listed when nothing can run commands, not usable, with the reason. - files: the file tools and read_document. - media: describe_image and transcribe_audio. - collections: list_collections and search_collection, usable only for a user with the collections feature. - subagents: spawn_agent. list_agents and /agent follow it. The registry gets two flags. External marks a tool that its own code builds for the turn (with the wrappers it needs); the registry lists it in its group and Hidden tells the builder what the chat switched off. Covers makes a group's switch also hide the add-on tool that replaces the built-in one. PUT .../builtins, the panel, the draft list and the fallbacks use the existing group code. A sub-agent follows the switches of its parent chat. The turn prompt is built from the final tool list: no file tool names, no skill writing hint, no work folder and no rules pointer when the chat cannot use files, no describe_image when the helper is off, and a commands block that says the person switched commands off. Assisted-by: Claude:claude-opus-5-5 * Queue the edited chat message after the first step starts The spec queued its message as soon as the chat showed a busy state. The turn is busy before its first model request starts, and that request takes every message that waits. With the chat's tools and skills, the first request now starts a few hundred milliseconds after the turn, so the message went with it, and the edit got 409 "already delivered" (5 of 6 local runs). Wait until the answer streams, as the Stop spec does: the request is running then, and the queued message waits for the next step. Assisted-by: Claude:claude-opus-5-5 * Give the reload chat spec a turn that outlasts the reload The spec reloaded the page during a 3 second turn and expected the page to show the turn still running. A reload of the coverage build on a busy runner takes 4 to 5 seconds (measured with a 6x CPU throttle), so the turn had ended before the page could show it. With the 30 second turn the page always showed it running, so the resumed state works and only the window was short. Add MOCK_LONG_STREAM_MEDIUM to the mock model, a turn of about 10 seconds, and use it in the spec with a 60 second limit. Assisted-by: Claude:claude-opus-5-5 * Wait for the chat page after the Home hand-off in the specs Send on Home changes the address inside a view transition, and the chat page mounts about 35 ms later. Specs that waited only for the address acted on Home in between: - The title spec pressed Ctrl+K at once, because the server names a chat after its first message as it arrives, so the title poll ends at once too. The key reached the page before the chat page listened, and the menu did not open (3 of 3 presses at 0 ms failed, all at 300 ms and later passed). - The Back spec went back before the chat page was drawn. Home was never closed, so its draft of the sent message stayed in the box. startChat now waits for the chat page, the title spec for the title in the chat header, and the Back spec for the chat page before Back. Assisted-by: Claude:claude-opus-5-5 * Accept a resolved user.json conflict in the quiet page specs The sidebar notes in the seen record of user.json the first day it labels Pages as new, and saves it from the version it read. All specs share the test server's one user, so tabs of other workers save the same record at the same time: one gets 409 and useSeen reads the file again and saves once more. Three tabs that open /app/files at once give 200, 409, 200. The quiet check of the shell pages counted the 409 as a problem. Ignore a 409 on PUT /api/chats/wellknown/user in that check. Assisted-by: Claude:claude-opus-5-5 * Check only the spec's own items in the Trash after the reload After Empty Trash and a reload, the spec expected the whole Trash to be empty. The Trash belongs to the test server's one user, and specs of other workers move files to it meanwhile, so CI saw "In the Trash: 3 files, 356 B." after the reload. The checks right after Empty Trash stay. After the reload, check that the server's Trash, the Trash tab and Files do not hold the spec's items, once each list has loaded. Assisted-by: Claude:claude-opus-5-5 * Show the new tool groups in the chat Tools menu The Tools menu lists the groups of built-in tools that the server sends. The five new ones need words and an order, and the shell row needs to say who runs the commands. Name the groups in all nine locales (Run commands, Your files, Images and audio, Collections, Sub-agents) and order the rows by what they reach: the shell and the files first, the triggers last. The shell row carries a chip with the provider (Sandbox or This machine) and a line under it that says the commands run isolated, or a warning when they run on this machine. When nothing can run commands the row is disabled with the words of the commands notice. A switched-off sub-agents group also turns /agent off in the composer. The specs cover the rows, the off and on requests, the bulk actions, the provider chip, the disabled reasons, the draft on Home and a real server round trip. Assisted-by: Claude:claude-opus-5-5 * Document the groups of built-in tools a chat can switch off The Tools menu section described only the web, triggers and saved agents groups. List every group with its tools, say which groups the server may leave out and why a row can be not usable, how Run commands covers both providers, what the model is told when a group is off, and that a sub-agent follows its parent's switches. Assisted-by: Claude:claude-opus-5-5 * Fix the host command consent spec for the 400 contract The consent route answers 400 for a body without a card id or with a decision other than once or deny, whoever asks. The spec still expected 403 for always from an administrator. Cover the malformed answers for an administrator and a user, each in their own chat, keep another user's chat at 404, and say it in the chats guide. Assisted-by: Claude:claude-sonnet-5-5 * Find the web tool group by name in the route spec The tools panel lists the shell group before the web group, so the spec that read the first entry compared the wrong group. Look the group up by name, which does not depend on the registry order. Assisted-by: Claude:claude-sonnet-5-5 * Give Skills its own sidebar item Skills work in every chat, but the sidebar filed the page under Agents and only agent skill chips linked to it. Add a Skills item right after Memory, show it in the legacy list too, and stop lighting up Agents on the skills pages. Assisted-by: Claude:claude-sonnet-5-5 * Add Manage skills to the skills picker and a /skills command The picker now ends with a Manage skills link, and /skills opens the Skills page from the composer on Home and in a chat. The picker keeps the height of its status line, so a refused change does not move the rows. Words are added in all nine locales. Assisted-by: Claude:claude-sonnet-5-5 * Say on the Skills page that every chat can use skills The page subtitle named only agents as users of skills. Say that all chats use them, translate it, and document where to manage skills: the sidebar item, the picker link and /skills. Assisted-by: Claude:claude-sonnet-5-5 * Keep Home on screen while its widgets load their strings A Home widget reads the pages strings, and i18next loads a namespace the first time a component asks for it, by suspending. Home had no Suspense boundary of its own, so the route's boundary hid all of Home (display: none) and showed the route fallback until the strings came. A widget shows only after home.json is read, so this happened after the composer was already drawn. Focus given to the composer in that window went nowhere, and a key typed then was lost. The slash menu spec failed in CI this way when another spec had written widgets to the shared home.json: the box was empty and the body had the focus. With widgets in home.json the composer was hidden right after it appeared on 5 of 5 loads, and the spec failed 1 or 2 times in 5. The examples and the widgets now have their own boundary with no fallback, so only they wait. A new spec holds the pages strings while it types in the composer; it fails without the change (the composer never takes the focus) and passes with it. Assisted-by: Claude:claude-opus-5-5 * Send the stubbed file change after the Files page read the folder The change feed spec answered the first stream request with the change at once. The app shell (the appearance sync) opens the shared stream when the app starts, before the Files page chunk has loaded, so on a slow runner the change reached a tab where nobody watched the folder yet and was dropped. The page then never read the folder again, and the spec failed on all three CI attempts. A CPU throttle of 8 makes it fail 3 of 3 times locally. On a real server the page's own first read would already show such a change. The stub now holds the change until the page has read the folder; the page subscribes in the same commit as that read, so the change arrives with a subscriber in place. The resume check is kept. Assisted-by: Claude:claude-opus-5-5 * Give the oversized 3D input as a sparse file, not a buffer The spec built a 32 MiB buffer and passed it to setInputFiles. Playwright sends a buffer to the browser over the protocol: that step alone took 5.5 s on an idle machine, and on the loaded CI runner the test hit the 30 s timeout on all three attempts. It now writes a sparse file of the same size to the test output folder and passes its path, which the local browser opens as is. The page still checks the size before it reads the file, so the toast and the no-request assertion are unchanged. The step takes milliseconds. Assisted-by: Claude:claude-opus-5-5 * Check the tool line, not the tool name, in the tool call journey The spec looked for /list_files/ in the transcript. The user's own message holds that name, so the check could pass before any tool ran. Once the model's call streamed, a second element matched, and the strict locator failed (2 matches on the first attempt, 5 on the retry). It now waits for the activity line that the tool run adds ("Listed /"), which only shows after the tool ran. Assisted-by: Claude:claude-opus-5-5 * Measure the proposal card and its neighbours in one frame The layout spec read the boxes of the message text, the card and the message actions with three boundingBox calls. They are viewport positions taken in three frames, and the transcript scrolls to its end while the card fills in. On the phone the actions were read after a scroll and came out 156 px above the bottom of the card. The three boxes now come from one page.evaluate, so they are read in the same frame. The order assertions are unchanged. Assisted-by: Claude:claude-opus-5-5 * Read the page fixture time when the page is served The task board fixture of the discovery and Pages specs set its change time once, when the file loaded. A worker loads it once and keeps running specs, so a spec that ran two minutes later read "Changed 2 minutes ago by the assistant" where it expects "just now". The change time is now a getter, read when the fixture is serialized into a response. Assisted-by: Claude:claude-opus-5-5 * Check a background process without /proc in the host test The alive helper read /proc/<pid>/stat, which macOS does not have, so it reported every process as dead and the spec failed there with "did not survive the suspend". Use signal 0 for existence and ps (or /proc where present) to tell a zombie from a live process. Assisted-by: Claude:claude-sonnet-5-5 * Keep the composer switch menus at their own height The base rule of the dropdown menu caps it at 320px and lets it scroll. The switch menus set their own height with a rule of the same weight in a lazy sheet, so which one won depended on the order the sheets loaded in. When the base rule won, the picker showed a few rows, and a click on a row at the edge of the list scrolled the list by a few pixels. Double the class on the switch menu rules so they win in any order, and add a test that loads the base rule last. Assisted-by: Claude:claude-sonnet-5-5 * Show what applies in the chat Network switch The Network row highlighted the choice stored with the chat, so a chat that asked for All public sites under an administrator limit of Off showed All highlighted although it was disabled and not in effect. Highlight the effective choice, the lower of the chat's choice and the limit, which the server already computes. When the chat asks for more than the limit, say so in one line instead of highlighting it. Add a line that the limit is set by the administrator, and tell the host network line that limits need the Sandbox add-on. Assisted-by: Claude:claude-sonnet-5-5 * Design the host components that add-ons reference The host UI names the Sandbox and the Web add-ons to place their server settings, and the chat's Network row links to the global Settings page. Describe a closed registry of host components that an add-on references by ID in its manifest, entitled by the existing host operation grants and tool shadows, so the controls live on the add-on's own page and a second add-on can reuse them with no host change. Assisted-by: Claude:claude-opus-5-5 * Let an add-on ask for host controls in its manifest An add-on had no way to say that a host-owned control belongs on its page, so the host UI placed those controls by add-on name. Add ui.controls, a list of IDs from a closed registry of host controls: network-policy (the network of the commands an add-on runs) and tool-providers (the rows of the built-in tools it replaces). The host decides which controls it draws: EntitledControls keeps a control only when the add-on holds its host operation grant and replaces its built-in tool. The field is optional, so every existing manifest stays valid. Assisted-by: Claude:claude-opus-5-5 * List the host controls an add-on is entitled to The apps list now carries ui.controls with the host controls the manifest declares and the host setting entitles the add-on to: the manager passes each add-on's host operation grants to the list. GET /api/chats/{id}/egress gains addon, the add-on that runs the chat's commands when it has the network-policy control, so the chat can draw its Network row from that add-on and link to its settings. applies and applies_reason keep their meaning: the policy holds with or without the control. The model and the 400 error no longer send an administrator to Settings > Tools for the limit. Assisted-by: Claude:claude-opus-5-5 * Draw the host controls on the add-on's own page The host UI placed the Sandbox network and the tool rows by add-on name (SERVER_SETTINGS), showed a read-only summary on the add-on's Settings tab, and linked the chat's Network row to the global Settings page. Add a registry of host controls keyed by the IDs of the contract. The Settings tab of an add-on draws, below its own sections, the controls the apps list gives it, in one form with the pending bar of the Settings page: network-policy reuses the network section and its setting rows, tool-providers the rows of the built-in tools the add-on replaces. The chat draws its Network row only when the server names an add-on with network-policy, and the administrator link opens that add-on's tab. The row itself is unchanged. The Settings page keeps the settings of features that exist without an add-on, points to the add-on where one serves them, and no longer lists the network settings. Its old address opens the add-on's tab. Assisted-by: Claude:claude-opus-5-5 * Ask for the host controls in the Sandbox and Web manifests The Sandbox asks for network-policy and tool-providers, so the network of its commands and the bash row are drawn on its own Settings tab and its chats get the Network row. The Web add-on asks for tool-providers only: it holds no egress grant. The manifest tests check that the default grants entitle each add-on to what it asks for. The proxy's refusal now sends the model to the Sandbox Settings tab instead of Settings > Tools. Assisted-by: Claude:claude-opus-5-5 * Keep the button of the network pointer row at its size The row used the wide layout of the tool rows, so its Open settings button stretched across the page. Use the plain setting row, which puts the button on the right like the other controls. Assisted-by: Claude:claude-opus-5-5 * Fix the add-on contract inventory and the apps Swagger fixture The contracts case inventory missed the six host control specs, so verify.sh rejected the Ginkgo report. List them. The Swagger spec test never saw `controls` in the emitted UIView JSON because its fixture left the omitempty field empty. Set it there. Assisted-by: Claude:claude-sonnet-5-5 * Add the host controls spec to the apps inventory The apps suite ran one more spec than scripts/addons/suites.json listed, so verify.sh rejected the apps report as an inventory mismatch. List the new spec in the report inventory and in the suite's aggregate case list. Every other Ginkgo suite was re-run and matches its inventory. Assisted-by: Claude:claude-sonnet-5-5 --------- Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-11-01 12:08:23 +01:00
mudler-agentandEttore Di Giacinto	c60a2e47d6	feat(backends): timestral.cpp forecast backend and /v1/forecast API (#12885) ... * feat(grpc): add a Forecast RPC for time-series models Add Forecast to the backend service with columnar request and response messages: flat row-major arrays, NaN for missing values, and one message per series so a caller can send series of different shapes in one call. The response keeps the request order. Wire the RPC through the server, client, embedded client, connection evicting client, in-flight tracking client and the base backend, which answers unimplemented. core/backend gains a Forecast helper that stamps the model identity and records a trace, like the other single-purpose RPCs. Assisted-by: Claude:claude-sonnet-5-5 go-vet * feat(config): add the forecast usecase FLAG_FORECAST marks a model as a time-series forecaster. The timestral-cpp backend declares it as its default usecase, so a config with that backend needs no extra lines. The backend is also listed as a non-text backend, so a forecast model is never offered as a chat model. It shows up in the gallery filter and in the capability list. Assisted-by: Claude:claude-sonnet-5-5 go-vet * feat(api): add POST /v1/forecast Take one or more numeric series and return quantile trajectories for the next prediction_length steps. Targets are [variates][steps] or a flat array for one variate; JSON null marks a missing value. Past and known covariates and categorical codes are checked for length before the backend sees them. Quantiles default to 0.1, 0.5 and 0.9 in the backend. The route is gated by a new default-on forecast API feature. The endpoint is listed in the swagger spec, the /api/instructions index, the UI model filter and the capability symbols. Assisted-by: Claude:claude-sonnet-5-5 go-vet * feat(backends): add the timestral-cpp forecast backend Wrap the timestral.cpp C ABI with purego, the way the other ggml backends do: libtimestral is built from a pinned commit with ggml embedded, and the Go binary loads it at start. Forecast validates the counts against the array lengths, groups the series by shape, makes one library call per group and puts the results back in the order the caller sent them. Quantiles default to 0.1, 0.5 and 0.9, and a level the model does not have is refused with the list of valid ones. A mutex guards the handle because the library is not reentrant. Request size is bounded by two model options. The unit specs use a fake library. The integration specs and a gRPC spec run against the 5M F32 model that test.sh downloads and checks against a fixed SHA-256. Assisted-by: Claude:claude-sonnet-5-5 go-vet * build(backends): register timestral-cpp in the build and the library Add the CPU image to the backend matrix for amd64 and arm64, plus an Apple silicon entry that is not built or tested yet. The image has no GPU keys because the engine runs on the CPU only. Wire the make targets, the path filter and the dependency bump bot. The library entries were generated with backendcatalog. The images are not published yet, so the release entries are unpinned until a maintainer runs --repin. The gallery reader test now counts the names in the library instead of a fixed number, so adding a backend does not break it. Assisted-by: Claude:claude-sonnet-5-5 go-vet * feat(gallery): add the Timestral forecast models Add Timestral 5M, 28M, 123M and 209M as F32 entries, each with its F16 build as a variant. Q8_0 and the K-quants are left out until the quality question is settled. Each entry runs on the timestral-cpp backend, on the CPU only. The weights use Mistral's Modified MIT licence, which bars use by companies above a monthly revenue threshold. The first paragraph after the summary says so, the licence file is installed next to the weights, and the revenue-restricted-license tag makes the entries filterable. The memory each build needs has not been measured, so no size is set. Assisted-by: Claude:claude-sonnet-5-5 * docs(forecast): document /v1/forecast and the Timestral models Add a feature page with the request and response shapes, the error cases, the model options and the licence note. List the engine in the README, the compatibility table and the backend categories. Assisted-by: Claude:claude-sonnet-5-5 --------- Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-24 22:47:04 +02:00
mudler-agentandEttore Di Giacinto	e45dc5e9f7	feat(addons): installable add-ons foundation and the Tables add-on (#12702) ... * feat(addons): add the add-on runtime, navigation and first-party add-ons Add the add-on runtime to LocalAI: manifests, the host operation broker, admin gating by account role, the add-on UI bundle builder and the SDK for add-on pages. Move the navigation to Chat, Models, Add-ons and Settings, with pinned add-ons in the sidebar and an Add-ons drawer. Ship the first-party add-ons (voice, face, fine-tuning, quantization) as packages for linux/amd64, linux/arm64 and darwin/arm64. Intel macOS (darwin/amd64) is enabled in the platform files. Squashed from the review branch. The tree is unchanged; the pre-squash history is kept on a local backup branch. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(gosec): tighten file modes and annotate false positives Write job state, configs and imported models with mode 0600, clean the paths that are read from disk, and mark the route tables, the header name and the retry jitter as reviewed false positives. Assisted-by: Claude Code:claude-sonnet-5-5 gosec Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): keep API keys off recovery mutations and same-site cookies POST /api/addons/dispatch forwarded the recovery and update mutations to the owner for any administrator key, which skipped the browser-session rule of their own routes. Refuse those actions there. Also apply the same-origin rule to the legacy token cookie, not only the session cookie, because the global CSRF check lets same-site requests through. Do not fail the start of LocalAI when an add-on library cannot be refreshed; log it and keep the installed add-ons up. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): refuse a reinstall that would migrate retained data unbacked An upgrade takes a verified backup before the new package migrates the data. A reinstall over retained data has no running generation to stop, so the new package would migrate it without one. Refuse a reinstall whose data format differs from the retained one. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): release the manager lock while a release is fetched Submit held the state lock across the catalog lookup and the package download, so the apps list and status reads stalled for the whole install. Serialize Submit and Reconcile with their own lock and drop the state lock around the fetch only. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): kill an add-on only after three missed health probes One slow 100 ms probe killed a busy add-on and dropped the work it was serving. Require three misses in a row. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): move settings under the operate hub The Operate hub already has Settings as a tab, so the sidebar no longer shows a separate Settings entry. The /app/settings route stays, and the hub tab still reaches it. The navigation test now expects Chat, Models and Add-ons as the primary items. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): accept same-host browser requests that carry no fetch metadata Browsers do not send Sec-Fetch-* headers over plain HTTP on an address that is not localhost, so every add-on request from a LAN address was refused as cross-site. Keep refusing a Sec-Fetch-Site value other than same-origin. When the header is absent, accept the request only if its Origin, or failing that its Referer, names the same host and port as the request. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): show the server's reason when an add-on request is refused A 403 or 400 answer showed a generic message and hid the reason in the response body, so a refusal by the browser guard could not be told from a missing permission. Show error.message from the body as the server sent it, and keep the generic text when the body has none. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): run the add-on owner in the normal binary The owner was only enabled when an embedding host set the add-on config, so a normal run answered every add-on request as unavailable. Add --addons-enabled (default true) and --addons-ownership-mode (default named-user, or local-owner; anything else stops startup). The node is its own owner: state lives under the data path, and the carrier uses loopback keys made at each start and kept in memory. --addons-enabled=false turns the owner off completely. Named-user mode needs the account database. With authentication off the owner stays off and a warning says how to use local-owner. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): keep the loopback owner address stable across restarts The single-node loopback transport took a new ephemeral port at each start. The owner stores its address in its identity file and refuses to start when the address differs, so every start after the first failed with "invalid addon owner configuration". Choose the port once, keep it in the add-on state root (loopback-port, mode 0600) and reuse it. If the port is taken at a later start, stop with an error that names the port and --addons-enabled=false, because another port would break the identity check. The keys and certificate stay in memory. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): let add-on pages keep the host modules they were written against A page written for the web UI imports React, the router, i18next and a few shared components. The add-on builder refuses a bundle that holds a host file, so moving such a page into an add-on meant rewriting it. Add a compat mode to the shared builder. It answers those imports from a directory of stand-ins that read the SDK, bundles the dynamic import of the page code into the one output file, and collects the style sheets the page imports. A relative import that is not in the add-on's own source resolves to the file at the same path under the compat directory. Widen the SDK for these pages, without a version change: more host components (spinner, modal, lightbox, model picker) and hooks (models, resources, voice profiles, operations, MCP client), a style on icons, the icon names the pages draw, sdk.navigation.openApp to open a page of the app (only addresses under /app, and the way back as the only state) and sdk.portal. The shell passes the query string to the page. The stand-ins cover the router (the tab in the path, the model in the query), the translator over the host locale files, the core API calls through sdk.core.fetch, and byte copies of the pure helpers, with tests that keep the copies equal to the originals. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): ship Studio and Talk as add-ons instead of built-in pages Studio and Talk were compiled into the web bundle and always listed in the Add-ons drawer. They are add-ons now, like Voice, Face, fine-tuning and quantization: a person installs and removes them, the drawer lists them only while they are installed, and they are packaged the same way. Move the pages, their components, hooks and helpers into addons/studio and addons/talk. Each has a manifest, a UI bundle, a release platform matrix and a catalog template, and is built by the compat mode of the shared builder. The core API stays where it is: the pages call it with the signed-in person's session. The history keeps its browser storage keys. The unit tests of the moved helpers move with them. In the core, drop the two pages, the Create group of the sidebar (Chat joins Home and Models in the old layout), the built-in catalog entries (the catalog is empty) and the talk locale namespace. The old addresses (/app/studio, /app/talk and the generator links) redirect to the add-on pages, with the model as ?model=. The Home /studio command is offered only while the Studio add-on is installed and opens it. The release workflow and its tests now list seven add-ons. Assisted-by: Claude Code:claude-sonnet-5-5 go node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): run the Studio and Talk specs against the add-on bundles The pages are add-ons now, so a spec that draws one installs the real built bundle first: the apps list names it and its module and style sheet are served from the build output, the way the add-on owner serves them. studio-addon.js holds that helper, and the Studio and Talk fixtures call it. Update the specs for the new addresses (/app/addons/studio/studio, the model as ?model=), the session permission the add-on route needs, the legacy redirects, the Home /studio command that appears only while the add-on is installed, and the sidebar without a Create group. Add specs for the add-ons themselves: every tab draws with no script error, the shell titles the page once, the old addresses say so when the add-on is not installed, a button of the page reaches a core page with the way back, and the drawer lists both only after they are installed. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): centre the icon of a built-in tile in its icon box The drawer drew the icon of a built-in tile (Studio's palette was one) in a bare 40px box. Only the placeholder box was laid out as a centred grid with the 20px size, so the real icon sat in the top-left corner at the size of the page text. Give the built-in tile the same classes as the placeholder box, and check in a browser that the icon is centred. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs(addons): describe Studio and Talk as add-ons Replace the built-in list (now empty) with the Studio and Talk add-on sections, the new targets of the old links, the Home command, and the seven packaged add-ons. Describe the SDK additions the pages use: components, hooks, icons, openApp and portal. Say on the Studio page that it is installed from the drawer. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): click the shell title in the Studio slash-key spec The add-on shell titles the page, and the Studio overview no longer draws a second title, but the generator tabs keep theirs, so a bare h1 locator matched two elements. Click the first, which is the shell's. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): read the loopback port file through os.Root The persisted port file was read with a path built from the add-on root, which gosec reports as a file inclusion through a variable. Open it through os.Root so the name cannot leave the root, even by a symbolic link, and require a regular file with no group or other access. A looser file now fails closed instead of being trusted. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): expect Settings under the Operate hub, not the sidebar Settings moved out of the sidebar into the Operate hub. The navigation spec still expected a sidebar link. Assert that the sidebar has no Settings entry, and that the Operate hub opens Settings as a tab. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): drop Settings from the 404 places Settings is a tab of the Operate hub and not a destination of its own, so the 404 page does not list it. Assert that it is absent. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): drive the media history and TTS specs through the Studio add-on The image, video, sound and speech generators are pages of the Studio add-on now, and the old addresses redirect to it. Install the add-on with the Studio fixtures, open its pages and use its compose card, so the specs run against the page that exists. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): count the first-run steps, not every list item The recommended models are a list inside the second step, so counting all list items read 3 only while the models were still loading. Count the step elements, which does not depend on the load order. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): fail the node refresh by state, not by call count The spec served the list to the first request and failed the rest. The first request is not always the one from the page, so the page could get the failure first and show no list. Serve the list until the page shows it, then fail the refreshes. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): default to local-owner when auth is off A server without authentication has no accounts, so named-user left the add-on owner off. Resolve the unset mode in one place: local-owner when authentication is off, named-user when it is on. A flag or environment value still wins, and an unknown value is still refused. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): register an official add-on library by default Add gallery/addons/index.yaml, a YAML index of the four first-party add-ons (fine-tuning, quantization, voice, face) in the library format. The entries come from addonpkg catalog over locally built linux/amd64 packages, so the digests pin those images. On the first start, when the owner has no sources.json, register the raw GitHub URL of that index on master as the "official" library and write the file. The file is then the record: a library the admin removes stays removed, and an existing file is never changed. --addons-official-library=false skips the default. Assisted-by: Claude Code:claude-sonnet-5-5 go node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs(addons): document the ownership default and the official library Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): list Studio and Talk in the first-party index Add the Studio and Talk entries to gallery/addons/index.yaml next to the four existing ones. They come from addonpkg catalog over locally built linux/amd64 packages, so the digests pin those images. Assisted-by: Claude Code:claude-sonnet-5-5 go node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): expect Studio and Talk in the official index The official index now lists six add-ons. Update the test that decodes it so it accepts Studio and Talk. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): add a real icon to each first-party add-on Each icon is a 256 by 256 PNG: a rounded square in one colour of the UI kit palette with a white outline glyph from the kit sprite. Fine-tuning uses sliders, quantization layers, voice waveform, face user, studio palette, talk phone and tables database. They were rendered with headless Chrome from the sprite paths. Assisted-by: Claude Code:claude-sonnet-5-5 chrome Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): ship a committed icon.png instead of the generated square The shared UI builder and the Tables builder copy ui/icon.png from the add-on source tree to the package unchanged when the file exists. They write the generated placeholder only when it does not. Add a fixture with an icon and a test for both cases. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): list the tables add-on in the first-party index Add the Tables entry to gallery/addons/index.yaml next to the six existing ones. It comes from addonpkg catalog over a locally built linux/amd64 package, so the digest pins that image. The official index test now expects seven add-ons. Assisted-by: Claude Code:claude-sonnet-5-5 go node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): make the Voice add-on one page with tabs The Voice add-on had four pages: recognition, library, create and speech, and the drawer listed two of them. It now has one page, "voice", and one drawer entry. Recognition, Library, Text to speech and From a recording are tabs of that page, drawn with the same tab strip the recognition page already used. The tab is the first path segment (voice/library), as in the Studio add-on, and the form that adds a voice is a view of the Library tab (voice/library/new). The old page names stay declared in the manifest. Each one redirects to its tab and keeps the query string, so saved links still open. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): send the old voice routes to the Voice tabs /app/voice, /app/voice-library and /app/voice-library/new now open the matching tab of the Voice add-on page while the add-on is installed and running, and the core pages otherwise. The moved-page helper compares the first segment of the target with the declared pages, so a target can name a tab. The docs and the navigation spec list the new addresses. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): cover the Voice tabs and the old page redirects The add-on unit tests render the tabs instead of the four pages. The Playwright spec checks the tab strip, the address of each tab, the missing Library tab for a member, and that the old page addresses of the add-on redirect to their tab with the query string kept. The migration table test follows the new targets. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(studio): keep the model menu inside a narrow screen The Home model picker sizes its menu from the viewport, and assumes the picker sits at the page gutter. In the Studio composer card the field is further in, so on a 390px screen the menu of the Sound expansion picker ended 7px past the right edge. Size the menu from the field it opens under instead. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): build the add-on bundles once for the whole run Each worker built the Studio and Talk bundles into the same dist directory, and the build empties that directory first. A worker that started, or restarted after a failure, removed the files another worker was serving, so a Studio spec saw "the style sheet of the add-on could not be loaded". Build once in a global setup before the workers start. A run without the setup still builds per worker. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): register the worker when the add-node spec says so The mocked roster added the worker on the third read of /api/nodes. The shell and the page both read it, and which one asks first depends on load, so the page could take the new worker as part of the roster it opened with and never show it as found. The worker now joins after the page shows it is listening. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): hold the gallery answer until the skeleton is seen The spec delayed the answer by a fixed 800 ms and then looked for the skeleton. The assertion polls at growing intervals, so on a slow machine it could step over that window. The answer now waits until the skeleton has been seen, and the spec then checks that the skeleton goes away. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): give the PDF tag the time the other PDF specs have pdf.js and its worker load on first use. The other specs in this file wait up to ten seconds for the result of that load. The first one waited five, and missed the tag on a busy machine. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): build the Face and Fine-tuning bundles in the global setup The navigation spec serves the built Fine-tuning bundle, but nothing built it, so two specs failed on a fresh checkout when the dist folder was missing. The spec also built the Face bundle in a hook that every worker runs, and a build rewrites the folder the other workers serve. The global setup now builds Studio, Talk, Face and Fine-tuning once, before the workers start. The spec builds the two bundles itself only when it runs without the global setup. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): fix the Home answers in the stale pin spec The spec asserts that the page shows no alert. The Home page raises its own alert when the model list fails or a staging operation has failed, and both answers come from the shared test server. A slow reply or a failed install left there counted as an alert about the pin. The spec now fixes both answers, so the alert count tells only about the pin. The assertions are unchanged. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): read the server address in the add-node spec from the page The spec expected the register address to carry port 8089, so it failed on any other port. It now expects the origin the page was opened from. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): give the navigation routes spec its own catalog No page is built in today, so the catalog is empty and the three specs that expect the studio and talk entries failed on the list and on the pin seed. They passed only while Studio and Talk were built in. The spec now sets a catalog of two ungated entries for each example and restores the old one afterwards. The assertions are unchanged. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(cli): report a failed write of the MLX hostfile The tunnel callback ignored the error of os.WriteFile and set the hostfile variable even when the file was not written. It now logs the error and leaves the variable alone. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): discard the close error of the held listener errcheck flags the deferred Close of the listener that keeps the port busy. The test has no use for that error, so the defer drops it explicitly. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): move the add-on tool tests to Ginkgo The fine-tuning, quantization and add-on host library tests used t.Errorf, t.Fatal and t.Run, which the forbidigo rule rejects. They now use Describe, It and Gomega, with a suite bootstrap in each package. The cases and the assertions are the same. The tables in the host client and the tool schema tests keep one entry per case. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): move the host operation tests to Ginkgo The hostops tests used the testing package calls that forbidigo rejects. They now run as Ginkgo specs under one suite bootstrap. Every check and every per-operation case is kept. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): move the pin store tests to Ginkgo The pin service, file store, sweep and shared store contract tests used the testing package calls that forbidigo rejects. The store contract is now a function that registers its specs for each store, so the file store, the memory-only store and the database store run the same cases. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): move the remaining add-on tests to Ginkgo The built-in catalog, tool arguments, loopback carrier, host operation contract, packaging, application, config, CLI and endpoint tests used the testing package calls that forbidigo rejects. They now run as Ginkgo specs. The endpoint tests also drop the unchecked Close errors that errcheck reports when the auth tag is on. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): wait for the targeted Reinstall button before the install click The "generic deletion blocks targeted install at confirmation" spec dispatches a popstate to set the target node, then clicks Reinstall with a script click. The click can run before the targeted page has rendered or while the button is still disabled, so no install request is sent and the poll on the install count times out. It failed 1 run in 10 with two workers and on every retry in CI. The spec now waits for the button to be enabled first. The assertions are unchanged. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore(addons): refresh the voice package digest in the first-party index The Voice add-on became one page with tabs and got a new icon, so the digest in gallery/addons/index.yaml no longer matched the build. Render the entry again with addonpkg catalog over a locally built linux/amd64 package; the main page is now voice. The other entries are unchanged. Assisted-by: Claude Code:claude-sonnet-5-5 go node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): wait for the target node's Reinstall button after the popstate The "generic deletion blocks targeted install at confirmation" spec dispatches a popstate to set the target node, then waits for a button named Reinstall and clicks it. Until React renders the targeted page, the page still draws the generic Reinstall button. It is enabled and has the same name, so the wait passed on it, the click started a global install, and no node install request was sent. The poll on the install count timed out with the confirm dialog still open. With the CPU throttled to 8x, the spec failed on 20 of 20 runs, and the button it clicked had the title "Reinstall". The spec now waits for the button titled "Reinstall on <node>". With the throttle it passes 30 of 30 runs. The assertions are unchanged. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): give the Studio hand-off spec a budget for the cold bundle load The "a result made through a hand-off records which result it came from" spec opens the Studio add-on straight from an address, twice. Each is a cold load: the page shows "Loading" until the add-on module has run, and the next assertion had the default five seconds. The load takes about two seconds on an idle run and seven to ten under CPU load, so the spec failed on a loaded machine with the hand-off note not yet drawn. With the CPU throttled to 6x and eight busy loops running, it failed on 21 of 30 runs. The first assertion after each load now waits up to 20 seconds, and the spec then passed 30 of 30 under the same load. The assertions are unchanged. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): carry the identity page styles in the Face add-on The Face page is drawn with the shared identity views, but its style sheet held only the clip and error rules. The page, card, model list, known-faces table, enrol sheet and distance scale classes had no rules. They looked right only when the host had already loaded the style sheet of its own Faces page, which a visit through the add-on drawer does not do. Opened that way, the cards had no gap, the model list showed bullets, and the install buttons ran into the names. The Face sheet now holds the same identity rules as the core sheet, so the page looks the same whichever way it is opened. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): scope the narrow-screen identity rules of the Voice add-on The rules for narrow screens (single-column pairs, smaller tool padding, compact rank rows) were copied into the Voice style sheet without their media query, so they applied at every width. They now sit in a max-width 720px query, and the reduced-motion rule in its own query, as in the core sheet. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): drop the repeated title from the Voice and Face pages The add-on shell already titles the page, so the recognition pages showed the name twice: "Voice" above "Voices", "Face" above "Faces". Keep the supporting line and the model picker, as the Studio page does, and remove the unused title strings. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): replace the Pinned sidebar section with an Apps entry The sidebar no longer lists pinned add-ons. It shows one Apps entry that opens the add-ons drawer, with a Show all link below it that opens the same drawer. Pins are used from the drawer, where a pinned tile keeps its pin and move buttons. The server pins API is unchanged. Remove the PinnedNav component and its locale title, rename the items.addons label to Apps, and update the e2e specs and the add-ons documentation to match. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): assert the single shell title on the Voice and Face pages The add-on shell now shows the one title, "Voice" or "Face", and the page no longer repeats it. The tests still looked for the h1 "Voices" and "Faces". Assert that the add-on page has exactly one h1 and that it is the shell title. The core pages keep their own "Voices" and "Faces" titles. Assisted-by: Claude Code:claude-sonnet-5-5 playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): drop the repeated title from the Voice Library and Text to speech pages The add-on shell already titles the page "Voice", so these two tabs showed a second heading below it ("Speech voices", "Speak text"). Keep the supporting line, as the Studio and recognition pages do, and remove the unused title strings. The e2e spec now checks that each tab has exactly one h1, the shell title. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): list pinned apps under Apps in the sidebar A pinned app now shows as an indented child of the Apps entry, with its icon and title, in pin order. Show all stays the last child. The pins come from GET /api/addons/pins through the navigation context, and the apps from the add-on registry, which the drawer feeds on every list read, so the links follow the drawer. A pin without an installed app is not drawn, nothing extra shows when no app is pinned, and the children are hidden when the sidebar is collapsed to the rail. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs(addons): say that pinned apps are listed under Apps in the sidebar The sidebar lists the pinned apps between Apps and Show all, in pin order. The pins section and the navigation section said that pins do not appear in the sidebar. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): remove the Show all link under Apps in the sidebar The link repeated the Apps entry, which opens the same drawer, and looked redundant when no app was pinned. Apps is now the only add-on entry; pinned apps still list under it in pin order. The link, its locale key and the style comment are gone, and the e2e spec and the add-ons docs now say that Apps is the only entry. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): show a first-run card on Apps when nothing is installed An administrator who opens Apps with no add-on installed now sees a short card with three steps: choose a library, browse and install, start the add-on. Its button opens the Browse tab. The card is hidden once an add-on is installed and for anyone who cannot install. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): hide the pinned apps in the icon rail When the window is narrower than 1024 px the sidebar shrinks to the icon rail by CSS alone, without the collapsed class. The pinned app links were still drawn there, with the indent meant for the full sidebar, so their icons sat to the right of the rail and were cut off. Hide the pinned app links in both rail states. The Apps entry keeps its centred icon. The e2e spec checks that the rail shows no pinned app and that every icon is centred at 1280 px collapsed and at 1000 px. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): stop the backend redirect test racing the installed list The test reads the redirected URL right after the path changes. The Backends page then loads the installed list and drops a selected backend that is not installed, so the backend parameter was sometimes already gone when the test read it. The mock server has no llama-cpp, and about 15 of 40 repeated runs failed on the backend assertion. The redirect itself was correct, including show_all. Serve the selected backend in the installed list, so the page keeps the selection and the assertions stay as strict as before. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): accept a same-host Referer or Origin for human confirmation Browsers send no Sec-Fetch-* headers over plain HTTP on an address that is not localhost, so creating a resource from a LAN address failed with "Creating a resource requires your own browser session". When Sec-Fetch-Site is present it must still be same-origin with an empty destination, as before. When it is absent, Origin or, failing that, Referer must name the request host and port. A destination or mode that is present must still be the expected one. This matches the rule in BrowserProtection. The credential, token cookie, API key and tool-context checks are unchanged. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): keep hub entries off the Apps route in the sidebar The Build hub still owned /app/addons through its legacy Add-ons tab, so the sidebar lit both Apps and Build on the Apps page. A tab whose gate is closed now owns no route when the sidebar asks, so only the Apps entry is active there. The hubs keep their own routes, and the legacy navigation still gives /app/addons to the Build hub. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): label the pin control on app tiles and in the detail view The pin was a small icon-only button and hard to find. Each app tile now shows a button reading Pin, or Pinned with a filled pin, and the detail view has the same control in its header. Both toggle the same stored pin. The Apps grid lists pinned apps first, in pin order, then the others. The sidebar list and the server pins API are unchanged. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): check the Voice From a recording tab has one title The Library and Text to speech tabs already assert that the add-on shell title is the only h1. Do the same for the From a recording tab, so a heading added there later fails the navigation spec. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): stop the add-on shell drawing a title over add-on pages The app decides what its page shows. The shell drew the add-on name as a header above every add-on page, so an add-on could not choose its own title. The shell now draws no header on a live, loading or frozen add-on page; each add-on titles its own pages. The unavailable and missing-page states keep the name, because no add-on page is drawn there. The runtime and shell specs now check that the shell adds no page title. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): let each add-on draw its own page title The shell no longer titles add-on pages, so each add-on draws one h1 per page. Voice titles its tabs "Voices" (Recognition and From a recording), "Speech voices" (Library) and "Speak text" (Text to speech); "Add a speech voice" already had its own. Face draws "Faces", Studio "Studio" and Talk "Talk". Tables promotes the database list and database name from h2 to h1. Fine-tuning and Quantization already drew a title. The add-on unit tests and the navigation and Studio specs now assert exactly one h1 per page, with the page's own title. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): add a kind to library entries and read unknown kinds tolerantly A library entry now has an optional kind: addon (the default when it is absent), backend or model. Every index written before kinds existed stays valid and its entries are add-ons. A reader keeps the entries of the kinds it knows and skips the others without reading their fields, so a newer kind never makes an older host refuse the whole index. The refresh result and the library state report how many entries were skipped and of which kinds, and the refresh logs a warning. A kind that is not a plain kind name still makes the index malformed. Search lists add-ons unless the caller asks for a kind, and Resolve returns add-ons only, so an add-on install cannot be pointed at a backend image. A library policy may name the signer with an anchored identityRegex instead of an exact identity, so builds from master and from tags both verify. An add-on install refuses a policy that carries one. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(backends): install a backend from a library entry A library can now list backends. An entry of kind backend names the image of the backend gallery by digest, and POST /backends/library/install installs it on the local node. The gallery stays the source of truth and its install path is the one used: the library resolves the entry, takes the signature policy from the library, pins the image by digest and calls the same install as the gallery. Installing a name that came from a gallery, another library or by hand is refused and changes nothing. The signature policy of a backend library names the signer with an anchored identityRegex, so images signed by the merge workflow on master and on tags both verify. A pattern that is not anchored is refused. An install is bounded by a size limit per kind: add-ons keep 512 MiB and a backend image defaults to 8 GiB, set with --addons-backend-max-bytes. The limit reads the manifest after the signature check and before any layer is fetched. The metadata of an installed backend records the library it came from. scripts/backends/cmd/backendcatalog renders entries from backend/index.yaml and pins them against the registry. The pilot is the CPU build of local-store, in gallery/backends/index.yaml. The library is registered on the first start only with --addons-backend-library, which is off by default. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): show the kind of library entries and install backends from them The Browse tab lists the entries of every kind this host reads and marks each one as an add-on or a backend. A backend installs through the backend route and not through the add-on lifecycle: the confirmation says that the backend gallery is not changed and that a backend runs native code, and a refusal of the host, such as a name installed from another origin, is shown with its own reason. The details give the kind, the alias and the signer pattern. The library editor can name the signer with a pattern. Without this a save would drop the pattern of a backend library. The Libraries tab reports the entries a refresh skipped because their kind is not read. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs(addons): describe backends in the library Document the entry kinds, how a reader skips a kind it does not know, the backend entry format, the signer identity pattern, the install route, the size limit of each kind, the pilot library and the command that renders it. The backend page links to it. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(mcp): install a backend from a library entry The backend library route installs a backend from an add-on library entry, and the admin REST surface and the MCP catalog must stay in step. Add the install_library_backend tool with an in-process and an HTTP client, the route map entry, the safety-rule listing and a skill recipe. Listing library entries stays on the add-on routes, which the assistant does not expose, so no list tool is added. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): carry the add-on library keys in every locale The add-on library and backend library strings existed only in English, so the other locales had no value for them. Add the missing keys of the addons namespace to de, es, id, it, ko, pt-BR and zh-CN. The values are the English text until a translation is contributed. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Revert "feat(mcp): install a backend from a library entry" This reverts commit 7493cff727. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Revert "docs(addons): describe backends in the library" This reverts commit 57816ed7e6. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Revert "feat(ui): show the kind of library entries and install backends from them" This reverts commit 49b2b151b7. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Revert "feat(backends): install a backend from a library entry" This reverts commit 1d8d3e3421. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * Revert "feat(addons): add a kind to library entries and read unknown kinds tolerantly" This reverts commit 484813f856. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): drop the backend library keys from the locales Revert the part of 39f77ae29 that added strings for backend libraries: the kind labels, the backend install texts, the signer identity pattern and the skipped entries notice. The add-on library keys stay. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): add an add-ons control to the chat header Add-on tools were on or off only through the /addons command, with a toast and no state a person could see. The chat header now has an Add-ons button in the style of the MCP dropdown. It lists each running add-on that ships tools with a switch, and a badge shows how many are on. The chat keeps the ids that are on (addonToolIds) next to the older boolean, and both are saved with the chat. A chat that has only the boolean means every add-on. The request sends addon_tools alone when every add-on is on, and adds addon_tool_ids when only some are. /addons turns all on or all off and uses the same state. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): let a chat request name the add-ons whose tools it uses addon_tools=true still offers the tools of every add-on the caller may use. A new addon_tool_ids metadata key, a comma-separated list of add-on ids, narrows that to some add-ons. The list counts only with the opt-in, and a list that names nothing offers no tool rather than all of them. The executor drops the tools of add-ons that are not listed and refuses a call to one as an unknown tool, so a model cannot reach a tool the person turned off. Both keys are consumed so they do not reach the backend. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): cover the per-add-on tool list and the chat header control Go tests cover the list form (the list filters the tools offered and the calls allowed) and the unchanged boolean form. Unit tests cover the chat add-on rules. Playwright specs cover the header control: listing only add-ons that ship tools, toggling each one, the count, /addons in step with the switches, and the request metadata. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): check each add-on page has one title when not enabled The shell names an add-on only where the add-on page cannot mount (not enabled, failed, no UI, bundle not loaded, no such page). Once the page mounts it draws its own title, including when it says it is not enabled on this server, and the shell draws none. The shell code already works this way; this change states the rule beside the unavailable state and pins it in the specs. The shell spec now expects exactly one title in the unavailable states. The navigation spec checks the Face page without a face model (the add-on title only) and the Face page of a stopped add-on (the shell title only). Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(addons): upgrade add-ons on their own when the setting is on A new auto_upgrade_addons setting (flag LOCALAI_AUTO_UPGRADE_ADDONS, default off) lets the owner upgrade an installed add-on when its library lists a newer version. It follows the backend setting: a startup flag, a runtime setting, and the same schedule as the backend update check (a first pass shortly after start, then every six hours). Each pass fetches the library of every installed add-on, and for each available update builds the same pinned upgrade command an administrator sends and runs it through the same lifecycle path. The channel rules, the signer identity, the digest checks, the SDK compatibility check and the backup before the upgrade all apply unchanged. The only difference is the gate: the owner acts on its own setting, so no administrator is asked. A refused or failed upgrade changes nothing and is logged; a candidate that failed is not retried until the library lists another digest. With the setting off, nothing runs and upgrades stay manual. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): add the Upgrade add-ons automatically toggle to Settings The toggle sits beside Upgrade backends automatically and is off by default. It saves the auto_upgrade_addons runtime setting. The English locale gets the label key, and the Settings test fixture lists the new field. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): cover the automatic upgrade setting and document it The setting defaults to off in the config, the runtime settings and the flag, and is read back from the saved settings file. With it on, a real owner upgrades an installed add-on through the manager; with it off, nothing changes, and a change of the setting at run time takes effect on the next pass. The manager specs show the upgrade goes through the same pinned path as the manual one, and that a corrupted package, a failed signature policy, an incompatible release, a missing update checker and a caller who is not an administrator each leave the installed version in place. The new specs are added to the lifecycle case inventory. The add-ons page describes the setting, its schedule and its default. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * refactor(addons): name the library entry identifier id The library entry of an add-on used addonId while the shared envelope and the model entries use id. Release now has the field ID, written as id, and every writer emits it: the index encoder, the package workflow templates, the catalog generator and gallery/addons/index.yaml. The digests in the index did not change. DecodeIndex still reads addonId as a legacy alias for one release, so an existing library and a library served by an older tool keep working. An entry that holds both names is refused. The Apps UI reads id and falls back to addonId. Assisted-by: Claude Code:claude-sonnet-5-5 go node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(addons): use the id field in the auto-upgrade test release The auto-upgrade test arrived after the identifier rename and still named the release field AddonID, so the application tests did not build. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(addons): annotate the auto-upgrade actor label for gosec The actor label looks like a credential to gosec G101. It is a fixed name for upgrades the owner starts itself. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(schema): persist the per-chat add-on tool fields The chat header persists addonTools and addonToolIds, but Conversation had no fields for them, so the server dropped them on save. Add both and update the generated swagger files by hand because swag is not installed here. Assisted-by: Claude Code:claude-sonnet-5-5 go Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): count the add-ons auto-upgrade setting in the settings groups The setting sits in the Backends and galleries group, once. The page now lists 54 settings plus the three branding images, so the spec expects 57. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): pause the fake clock ahead of now in the clear-all spec The fake clock runs from install, so pausing at the current time failed with 'Cannot fast-forward to the past' on a slow runner. The save itself is synchronous and was not lost. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(ui): hold the Studio capabilities reply longer in the skeleton spec 800 ms was shorter than a slow runner needs to render the page, so the skeleton was gone before the check. The product shows it. Assisted-by: Claude Code:claude-sonnet-5-5 node playwright Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(i18n): add the French add-ons locale Master added the fr locale after this branch forked. The add-ons unit test requires every locale to carry the add-ons namespace, so fr needs its own addons.json. Assisted-by: Claude Code:claude-sonnet-5-5 node Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-22 17:57:19 +02:00
mudler-agentandEttore Di Giacinto	dce91ba7aa	fix(backends): prefer installed development builds (#12782) ... * fix(ci): capture and surface ginkgo errors in run-coverage.sh The coverage script ran ginkgo per test root with `|| fail=1`, which swallowed the exit code. Combined with `make --output-sync=target`, which buffers all output and flushes at the end, the actual ginkgo error was invisible in interleaved CI logs. Every test suite passed but the script still exited 1, and the error message was buried in 22 000 lines of buffered output. Each ginkgo invocation now captures combined stdout+stderr to a per-root log file. On failure the last 80 lines are printed to stderr so the error is visible in the CI log. On success the log is replayed to stdout so test output remains visible. The build flags (--tags, --coverpkg) are quoted correctly so values containing spaces (e.g. --tags="debug auth") are preserved through eval. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ci): create output dir before writing ginkgo logs The previous commit added per-root log files but forgot to create the output directory before the first ginkgo invocation. On CI the coverage/ dir did not exist yet, so the shell redirect to cover-pkg.out.log failed with "No such file or directory" before ginkgo even ran. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(core/http): add canvas preview route to auth coverage allowlist Commit 9ad78381d8 added GET /api/canvas/preview/:id as a public route (no auth — the random id is the capability, an iframe cannot send API key headers) but did not add it to expectedPublicRoutes in the route auth coverage test. The test flagged the route as an anonymous-access leak, failing the entire ./core ginkgo suite on every CI run since. The POST /api/canvas/preview endpoint is auth-gated (returns 401 to anonymous requests) and is intentionally absent from the allowlist. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(backends): prefer installed development builds Prefer compatible runnable development builds for generic aliases when the preference is enabled. Keep concrete names and disabled selection unchanged. Pass worker-local configuration into resolver state before registration. Synchronize saved frontend settings at startup and require a restart for resolver changes instead of switching registered paths during runtime. Add selection, root-shadowing, configuration, and restart-boundary tests. Document fallback behavior and worker-local activation. Assisted-by: Codex:gpt-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(settings): isolate backend restart notices A development preference save must not restart the watchdog. Separate its server restart notification from the legacy reconfiguration signal. Preserve existing response fields and genuine watchdog updates. Exercise the real settings endpoint for both preference values, repeated saves, and mixed watchdog updates. Verify resolver state and registered paths remain unchanged until startup. Attribution correction for commit 85c46d2fac: its Assisted-by: Codex:gpt-5 trailer is incorrect. The actual agent was nib using model gpt-6.1-sol. Preserve that commit as task history. Assisted-by: nib:gpt-6.1-sol Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-19 22:58:38 +02:00
4 people	36307005a9	feat(chat-history): server-side chat history with tombstones (follow-up to #9902) (#12663) ... * feat(schema): add Conversation schema for chat history (#9432) Persist WebUI chat conversations server-side so browser refresh, private windows, or device changes preserve user history (issue #9432). This commit adds the on-disk representation: - Conversation holds id, name, model, opaque history, MCP / sampling settings, and timestamps. - ConversationsFile is the per-user list serialised to JSON. History is stored as json.RawMessage so the server stays agnostic to React message shapes (user / assistant / thinking / tool_call / tool_result mixed with text / image_url / audio_url / file attachments). Assisted-by: Claude:claude-opus-4-7 Signed-off-by: TLoE419 <tloemizuchizu@gmail.com> * feat(chathistory): add file-based per-user conversation store (#9432) File-backed persister for chat history. Each user's conversations live under {baseDir}/{userID}/conversations.json (anonymous/ when auth is disabled). Design choices: - In-memory cache backed by sync.Mutex so concurrent saves don't interleave writes. - Atomic write via tmp file + os.Rename so a crash mid-save never leaves a corrupted history. - ID validation regex blocks path-traversal payloads at the store boundary, in addition to the auth context constraint. Tests cover round-trip persistence across instances, user isolation, unsafe-ID rejection, bulk replace migration, and the anonymous fallback path. Assisted-by: Claude:claude-opus-4-7 Signed-off-by: TLoE419 <tloemizuchizu@gmail.com> * feat(api): expose chat conversation CRUD endpoints (#9432) Wire the chathistory store into the HTTP layer. New endpoints under /api/conversations: - GET /api/conversations list - POST /api/conversations upsert (id in body) - DELETE /api/conversations delete all - PUT /api/conversations/bulk replace entire set (localStorage migration) - GET /api/conversations/:id fetch one - PUT /api/conversations/:id upsert (id in path; path id wins over body) - DELETE /api/conversations/:id delete one All endpoints scope queries to the authenticated user via getUserID(c) - callers cannot impersonate other users by passing a user_id in the body. The endpoints are gated behind the new chat_history feature permission (default ON in APIFeatures), registered through the existing RouteFeatureRegistry so the unified feature middleware picks them up automatically. Application.ChatHistoryStore() returns nil when DisableWebUI is set or no persistence path is configured, in which case route registration is skipped entirely rather than registering handlers that always 503. Also adds the chat-history instruction entry and Swagger tag so the endpoint surfaces in /api/instructions and /swagger. Assisted-by: Claude:claude-opus-4-7 Signed-off-by: TLoE419 <tloemizuchizu@gmail.com> * feat(react-ui): sync chat history to server with localStorage fallback (#9432) useChat now probes /api/conversations on mount. When the endpoint responds 200, the server becomes the authoritative source: the hook merges remote conversations into local state and pushes per-chat PUT updates on each debounced save. When the endpoint 404s - older LocalAI deploys or the feature disabled - the hook silently keeps the old localStorage-only behaviour, so the change is backward-compatible. Migration: when the server is reachable but empty and the browser has local conversations with history, the hook fires a single PUT /api/conversations/bulk to seed the server. The atomic bulk endpoint avoids partial-upload states where a retry would skip already-uploaded entries. Delete operations propagate to the server when it's available; failures are silently swallowed so the local UI stays responsive on transient network errors. Assisted-by: Claude:claude-opus-4-7 Signed-off-by: TLoE419 <tloemizuchizu@gmail.com> * test(chathistory): rewrite tests in Ginkgo/Gomega The initial test file used the stdlib testing package (t.Fatalf / t.Errorf), which is forbidden by the forbidigo linter rule in .golangci.yml — LocalAI mandates Ginkgo/Gomega for all Go tests (see .agents/coding-style.md). Restructured the same six scenarios into Describe / Context / It blocks, with a chathistory_suite_test.go bootstrap that registers the Ginkgo fail handler. Test coverage is unchanged: basic CRUD round-trip, cross- instance persistence, user isolation, unsafe ID rejection (now via DescribeTable), bulk replace overwrite, and the anonymous fallback path. Assisted-by: Claude:claude-opus-4-7 Signed-off-by: TLoE419 <tloemizuchizu@gmail.com> * refactor(chathistory): switch store to GORM-backed authDB Addresses mudler's review on #9902. Chat history was the only per-user state in LocalAI sitting in a separate filesystem store instead of the shared GORM database — AgentStore and JobStore both live there. Move it onto Application.authDB so per-user state stays in one place. * ConversationRecord (GORM model) with composite primary key (user_id, conv_id) — React mints IDs locally with no global coordination so per-user uniqueness is the natural shape. * Store.New(db *gorm.DB) replaces Store.New(baseDir string). Save / Get / List / Delete / DeleteAll / ReplaceAll keep their signatures so HTTP handlers and the React UI are untouched. * ReplaceAll runs in a transaction with an Unscoped() pre-clear so a future retention sweep cannot resurrect a stale ID on re-upload. * gorm.DeletedAt in the schema means a future retention pruner is a one-line query — addresses the second half of the review (unbounded in-memory map). The map is gone: reads go straight to the DB. Trade-off worth flagging in the PR reply: chat history persistence now requires auth.Enabled = true (that's what initialises Application.authDB). With auth disabled the React UI transparently falls back to localStorage, the same path the original PR took on 404. Consistent with AgentStore / JobStore; if a "chat history always on" mode is preferred we can pull DB init out of auth.InitDB into a shared helper. Assisted-by: Claude:claude-opus-4-7 [Read Edit Bash] Signed-off-by: TLoE419 <tloemizuchizu@gmail.com> * refactor(schema): drop now-unused ConversationsFile envelope ConversationsFile was the JSON wrapper for the old file-based chat history store — a {Conversations: [...], UpdatedAt: ...} pair serialised to disk under {baseDir}/{userID}/conversations.json. The previous commit replaced that store with a GORM-backed one that writes rows directly as ConversationRecord, leaving the envelope without a consumer. Assisted-by: Claude:claude-opus-4-7 [Read Edit Bash] Signed-off-by: TLoE419 <tloemizuchizu@gmail.com> * test(chathistory): switch tests to testutil.SetupTestDB Mirror the AgentStore test setup: spin up a PostgreSQL 16 container via testutil.SetupTestDB(), construct a Store on top of it, exercise CRUD round-trips. The previous file-based fixtures (TempDir, filepath.Join checks) are no longer meaningful now that the store talks GORM. Tests are skipped on macOS (testcontainers needs Docker) and run on CI, same as the other GORM-backed suites in the repo. Also reframed two cases: * "unsafe IDs" → "malformed IDs" — the DB-backed store doesn't care about path traversal, but idRegex still rejects whitespace / control characters so IDs stay safe in logs and HTTP responses. * "anonymous user" no longer pins a directory path (no directory exists anymore); instead it checks the round-trip plus the per-user isolation guarantee. Added a DeleteAll round-trip to cover the bulk-clear endpoint. Assisted-by: Claude:claude-opus-4-7 [Read Edit Bash] Signed-off-by: TLoE419 <tloemizuchizu@gmail.com> * test(chathistory): cover the conversation HTTP handlers The seven /api/conversations handlers had no tests. Inject the *chathistory.Store into the handlers instead of the whole Application and move the route table into RegisterChatHistoryRoutes, so a spec can drive the real registration and middleware chain (auth.Middleware plus the chat_history feature gate) against an in-memory SQLite auth database. The new specs cover 401 without credentials on every route, 403 when an admin revoked chat_history, the CRUD round trip, path id over body id, 400 on malformed ids and bodies, per-user isolation, bulk replace and delete-all, and that a nil store registers nothing (the UI then falls back to localStorage). The handlers' nil-store branches are gone: the routes are only registered with a store, so those branches could never run. Assisted-by: Claude Code:claude-opus-5-5 Signed-off-by: Stefan Walcz <stefan.walcz@walcz.de> * fix(chathistory): make deletions final with tombstones and versions A conversation deleted on one device came back as soon as a client with a stale copy wrote it again: Delete soft-deleted the row, and the next Save went through GORM's upsert, which cleared deleted_at and resurrected it for every device. The lookup that recovered CreatedAt also ran outside any transaction, so concurrent writers raced on it. The server is now authoritative over deletions: * Delete turns the row into a tombstone: content dropped, deleted_at set. It is idempotent, and deleting an id the server has not seen yet still records a tombstone, so a delete that overtakes the UI's in-flight debounced save of the same chat still wins. * Save runs in a transaction, refuses a tombstoned id with ErrDeleted (HTTP 410 Gone), and bumps the version with a single UPDATE ... SET version = version + 1 guarded by deleted_at IS NULL, so concurrent writers each count and a delete committed between the lookup and the write still wins. Two concurrent first writes of the same id retry once (ON CONFLICT DO NOTHING, then update). * Every row carries a server-assigned version (1 on the first save, +1 per write including the delete), returned to the client. A non-zero version tells the UI the server has seen a conversation, which is how it can tell a never-synced draft from a server-side deletion. * GET /api/conversations returns the tombstones under "deleted". * DeleteAll and ReplaceAll leave tombstones too. ReplaceAll no longer hard-deletes everything first; it upserts the payload, skips ids that were deleted (reported in "skipped"), and tombstones the rest. * createdAt/updatedAt are stored as the client sends them (only zero values are stamped), so clients compare updatedAt from their own clocks rather than against the server's. The store specs now run against PostgreSQL (testcontainers) and, with the auth tag, SQLite through auth.InitDB, since both back the auth database. The API specs cover a stale client reconnecting after a remote deletion. Swagger is regenerated with swag v1.16.6 (json.RawMessage fields get swaggertype), and the feature is documented, including that it needs user auth. Assisted-by: Claude Code:claude-opus-5-5 Signed-off-by: Stefan Walcz <stefan.walcz@walcz.de> * fix(react-ui): let the server decide which chats were deleted mergeRemoteAndLocal kept every chat that was only in the browser and let the server copy win for every shared id. A stale browser therefore uploaded chats deleted on another device as if they were new drafts, and a message sent while the list request was still in flight was replaced by the older server copy and then lost from localStorage too. The first load now reconciles with the server's tombstones and versions: * a chat the server lists under "deleted" is dropped locally; * a chat on both sides keeps the copy with the newer updatedAt, so an edit made during the load wins and is uploaded afterwards; * a chat only in the browser is uploaded only if it was never synced (no version). With a version the server has seen it and no longer has it, so it is dropped instead of resurrected (this also keeps one account's cached chats out of another account on a shared browser); * deletes made while the list is loading are queued and sent once it arrives. Saves record the version the server returns, and a 410 on save drops the chat locally instead of retrying it forever. Chats received from the server count as already sent, so adopting a server copy no longer echoes it straight back. The bulk localStorage migration is gone: it replaced the whole server set from a possibly stale browser, and the per-chat saves already upload never-synced drafts through the tombstone check. The Playwright spec drives each case against a stubbed /api/conversations that follows the server contract. Assisted-by: Claude Code:claude-opus-5-5 Signed-off-by: Stefan Walcz <stefan.walcz@walcz.de> * fix(chathistory): keep the LocalAI Assistant toggle on synced chats master's createNewChat sets localaiAssistant, but serializeChat did not send it and schema.Conversation had no field for it, so a chat with the Assistant mode on came back with it off after a reload from the server (and the local cache dropped it too). Add the field on both sides. The other fields createNewChat sets were already covered; a handler spec now round-trips all of them, and the UI spec checks the toggle on the way up and in the local cache. Assisted-by: Claude Code:claude-opus-5-5 Signed-off-by: Stefan Walcz <stefan.walcz@walcz.de> * fix(react-ui): scope chat cache per account, retry unconfirmed deletes Two gaps in the client half of server-side chat history, both found in review: The browser cache was one shared localStorage key. A draft that one account could not upload (a failed first save) stayed in it, and after that account signed out and another signed in, the reconcile kept the unversioned draft and the save pass uploaded it into the second account. The cache now lives under localai_chats_data:<user id>, Chat remounts its state when the account changes, and nothing is synced without a signed-in account. The shared key is left for browser-only use without auth and is not uploaded when signed in, because nothing tells which account it belongs to. Deletes were dropped locally and their DELETE errors swallowed; only deletes made while the first list was loading were queued, and not persisted. A delete made while the server was unreachable therefore came back with the next list. Deletes (single, delete all, and deletes on the Home page) are now kept as pending deletes in the account's cache until the server confirms them, override the server list in the reconcile, and are sent again on the next change, on the online event and on the next load. While a delete-all is pending no conversation is uploaded, so a chat started after it is not caught by it. The Playwright specs replay both review scenarios (account switch after a failed first save; delete and delete-all while offline) with counter-checks that the same account keeps and uploads its own draft and that a confirmed delete is neither re-sent nor resurrected. Assisted-by: Claude Code:claude-opus-5-5 Signed-off-by: Stefan Walcz <stefan.walcz@walcz.de> * fix(react-ui): persist cleared chats immediately Navigation cancels the debounced cache write and can restore unsynced chats after Clear all. Save the cleared list with its pending deletion before returning, and cover immediate navigation in the browser spec. Assisted-by: Codex:GPT-6 Signed-off-by: Stefan Walcz <stefan.walcz@walcz.de> * test(react-ui): make the chat history sync specs deterministic - "clear all persists before navigation" opened a single chat, and the menu offers no Clear all for one chat, so the spec timed out in every run. It now opens two unsynced chats and checks that neither survives. Reverting 03cd9c115's save in clearAllChats turns it red again. - "a delete the server confirmed is not sent again" reopened the page as soon as the server had the request, before the page had the answer; the unconfirmed delete was then rightly sent again (4 of 5 runs). It now waits until the cache holds no pending delete. Docs-Not-Needed: test-only change. Assisted-by: Claude Code:claude-opus-5-5 Signed-off-by: Stefan Walcz <stefan.walcz@walcz.de> * fix(chathistory): keep the chat's reasoning effort on the server reasoningEffort (#12700, #12716) is part of the chat the UI syncs since the master merge, but schema.Conversation had no field for it, so the server dropped it and a reload from the server reset the setting. Add the field. The handler spec that round-trips every chat field now includes reasoningEffort. A schema test reads serializeChat in useChat.js and fails for any key the UI sends that schema.Conversation cannot hold, so the next such drift shows up as a failing test rather than in review (it reports reasoningEffort on the previous head). Swagger and docs updated. Assisted-by: Claude Code:claude-opus-5-5 Signed-off-by: Stefan Walcz <stefan.walcz@walcz.de> * test(chat): use Ginkgo for the schema drift guard The new drift guard uses testing methods forbidden by repository lint. Keep its field coverage and fail-fast parser checks in Ginkgo/Gomega. Assisted-by: nib:gpt-6.1-sol --------- Signed-off-by: TLoE419 <tloemizuchizu@gmail.com> Signed-off-by: Stefan Walcz <stefan.walcz@walcz.de> Co-authored-by: TLoE419 <tloemizuchizu@gmail.com> Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com> Co-authored-by: mudler-agent <mudler-bot@c3os.io>	2026-10-19 22:52:07 +02:00
mudler-agentandEttore Di Giacinto	ae8092e1c6	feat(sound): expand Simple prompts with chat (#12772) ... * feat(sound): expand simple prompts with chat models Allow callers to select an installed chat model for sound prompt expansion. Validate its JSON before audio generation and preserve explicit instrumental and language controls. Keep backend defaults when expansion is omitted, including native Python ACE-Step sampling. Assisted-by: Codex:gpt-6 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(sound): map expansion by loaded audio family Map neutral expanded fields where the resolved audio family is known. Keep unmarked requests unchanged and reject unsupported expansion before inference. Clear YuE2 and HeartMuLa transcripts to prevent lyric fallback. Assisted-by: Codex:gpt-6.1-sol Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(sound): choose a prompt expansion model Require an explicit Simple choice and remember it separately from chat. Clear unavailable choices instead of silently switching language models. Preserve the choice in history and restore it when editing a past result. Keep Backend default and Advanced requests compatible with existing APIs. Document family support and align the published optional request schema. Verify selection, request fields, persistence and history in the browser. Assisted-by: Codex:gpt-6.1-sol Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(sound): authorize both generation models Check the resolved audio model and explicit expansion model before loading either. Permission lookup failures stop generation. Assisted-by: Codex:gpt-6 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(sound): enforce JSON during prompt expansion Constrain expansion output with the existing JSON grammar before decoding. Require a llama-cpp chat model so unsupported backends cannot ignore it. Keep strict field validation and fail before generating audio. Assisted-by: Codex:gpt-6 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-19 18:16:20 +02:00
3 people	e715af1def	feat(llama-cpp): add slot prompt-cache save and restore API (#12730) ... * feat(llama-cpp): add slot prompt-cache save and restore API Add POST /v1/slots/save and POST /v1/slots/restore. They write one llama.cpp slot's prompt cache (KV state) to a file and load it back, so a client can warm a long system prompt in a few hundred milliseconds instead of re-running prefill. Refs #12685. The new SlotCache gRPC call posts the same slot save and restore tasks the llama-server /slots endpoint uses. It is off unless the model sets the new slot_save_path option, so existing models are not affected. The /slots monitoring endpoint is not needed here: the backend has no HTTP server. The file name comes from the caller, so it is checked twice, in Go and again in the backend. It must be one path component: no empty name, "/", "\", "..", NUL, control characters or absolute path. The resolved path, with symbolic links followed, must be a direct child of slot_save_path. The routes are admin only because they write files on the backend host. Assisted-by: Claude:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(llama-cpp): answer 501 when a backend lacks slot cache A backend that does not implement the SlotCache rpc returns gRPC UNIMPLEMENTED. The API passed that text through, so the client saw "unknown method SlotCache". Map the status to a typed error in core/backend and return 501 with a fixed message instead. Every other error path is unchanged. Make the SlotCache comments in backend.proto backend-neutral: slot is the backend's state slot, filename must stay inside the save directory, and an empty model_identity means the backend does not check it. Only comments change, no field or name. Add tests that the identity sent is the configured model, and for the error mapping. Assisted-by: Claude:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(llama-cpp): issue slot cache IDs on the server The slot save and restore API let the client pick the file name. Restore could load any file in slot_save_path, and the symlink check ran before llama.cpp opened the file. The backend now issues the name. A save returns a random 128-bit ID (32 lowercase hex characters, read from /dev/urandom) and never overwrites an existing file. Files live in a 0700 directory per model, named from the SHA-256 of the model identity, with a manifest.json that records each save. The manifest is written atomically and a corrupt one stops both save and restore. A restore takes the ID, finds it in the manifest, refuses an entry of another model, builds the path from the entry and checks with lstat that it is a plain file of the recorded size. The directory mode is the main defence against another local user, since llama.cpp still opens the path itself. Each model keeps at most 32 saves and the oldest is deleted first. A file over 8 GiB is deleted after the save, because llama.cpp does not report the size before it writes. Responses carry no host path. The HTTP bodies change: save takes {model, slot}, restore takes {model, slot, id}, and an unknown ID returns 404. In the proto, SlotCacheRequest.filename and SlotCacheResponse.filename become id. SlotCache no longer calls checkModelIdentity, which reads a field this request does not have. Assisted-by: Claude:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(llama-cpp): copy slot_cache files in prepare.sh grpc-server.cpp includes slot_cache.h, but prepare.sh copies the helper headers into llama.cpp/tools/grpc-server/ one by one and skipped it. The llama-cpp, turboquant and bonsai builds share this step, so they all failed with "slot_cache.h: No such file or directory". Copy slot_cache.h and slot_cache_test.cpp, which CMakeLists.txt builds when LLAMA_GRPC_BUILD_TESTS is on. Assisted-by: Claude:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore(deps): bump LocalAGI to pick up the scheduler start/stop fix LocalAGI's scheduler could hit a nil pointer in run when Start and Stop raced. That made the core/services/agentpool tests panic at random and turned tests-linux red. Move to mudler/LocalAGI@7aa0c8fea0, the merge of mudler/LocalAGI#519 ("make Start and Stop safe to call concurrently"). The new module version also adds a few indirect requirements to go.mod and go.sum. Assisted-by: Claude:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@users.noreply.github.com>	2026-10-17 18:23:02 +02:00
mudler-agentandEttore Di Giacinto	e28d5009d9	stablediffusion-ggml: run MiniMax-H3 video with references and add its gallery entries (#12668) ... * feat(stablediffusion-ggml): run MiniMax-H3 video with reference conditioning Bump stable-diffusion.cpp to 2988060 and follow its tiling rename (tile_size_x/y -> tile_size_w/h, rel_size_x/y -> rel_size_w/h). GenerateVideo now passes params["ref_image"] and the request audio to the MiniMax-H3 Ref2VA inputs (ref_images, ref_audios). The audio must be a WAV file. params["ref_video"] is refused, since the proto cannot carry a frame sequence yet. stable-diffusion.cpp cannot tell the FL2VA and REF2VA DiTs apart, and a reference input on FL2VA renders a wrong video without an error. The backend now reads the video_partition option, or the model file name, and fails early on that mix. Keyframes plus references are refused too. Add gallery entries for the FL2VA and REF2VA Q4_K_M sets. They do not enable any bf16 dequantisation. Assisted-by: Claude:claude-sonnet-5-5 go vet gofmt * feat(stablediffusion-ggml): take several reference images, videos and audio clips Phase 1 passed one reference image and one reference audio clip to stable-diffusion.cpp and refused ref_video. Add repeated reference fields to GenerateVideoRequest (ref_images, ref_videos, ref_audios) so the MiniMax-H3 Ref2VA path can use as many references as upstream accepts. A reference video is a directory of frames with an fps and an optional paired audio clip. The /video endpoint stages the new inputs through the same path as start_image, end_image and audio. Frames of one reference video are written to a private directory under zero-padded names so the backend can read them in order. The input count and frame count per request are bounded. The backend merges the new fields with params.ref_image, params.ref_video and audio, which keep working. The FL2VA partition refuses any reference input, REF2VA accepts all of them, and keyframes stay exclusive with references. Reference audio is still WAV only. Assisted-by: Claude:claude-sonnet-5-5 [Bash] * feat(gallery): offer MiniMax-H3 by partition, DiT layout and format Name each MiniMax-H3 entry after its partition (fl2va or ref2va), DiT layout (unpruned or pruned) and format. Add the vllm.cpp builds that its model page lists: pruned Q8_0 (renders), pruned Q6_K, Q5_0, Q4_K, Q3_K and Q2_K (load only), and the pruned FL2VA NVFP4 file (loads only). The earlier names minimax-h3-fl2va-q4 and minimax-h3-ref2va-q4 stay as aliases. The two stable-diffusion.cpp entries become minimax-h3-fl2va-unpruned-q4-ggml and minimax-h3-ref2va-pruned-q4-ggml, and the latter now lists video input. Fix two bad sources. The stable-diffusion.cpp entries took the VAEs from a path that does not exist in the GGUF repository; they now use the Comfy-Org copies. The vllm.cpp entries paired the fp16 video VAE weights with a config from a different directory; they now use the official weights and config from FL2VA/video_vae/source/. Drop video_dequant_bf16 from every vllm.cpp entry. Expanding a quantised DiT to bf16 is the suspected cause of a host out-of-memory error, so the quantised weights stay resident. The model page records its renders with the expansion on, and each description says so. Assisted-by: Claude:claude-sonnet-5-5 [Bash] * feat(gallery): add pruned FL2VA Q4 MiniMax-H3 for sd.cpp Add minimax-h3-fl2va-pruned-q4-ggml, the 11.4 GB pruned FL2VA counterpart of the existing pruned REF2VA entry. Assisted-by: Claude Code:claude-sonnet-5-5 [rc] --------- Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-13 17:58:21 +02:00
mudler-agentandEttore Di Giacinto	0177e4988c	feat(images): add ESRGAN upscaling and Studio workflow (#12620) ... * feat(images): add upscale model capabilities * fix(config): keep SD GGML upscale opt-in * feat(stablediffusion): add lazy ESRGAN upscale bridge Keep native image ownership in C++ and serialize context creation, inference and teardown. Use upscale_scale in backend options as the explicit upscale-only marker because ModelOptions does not carry usecases. Assisted-by: Codex:gpt-5 * fix(stablediffusion): validate ESRGAN settings * fix(upscale): forward typed settings and contain resolved model paths Typed settings replace legacy options without duplicates. Resolve model and trusted root symlinks before caching the model path, rejecting missing files and escapes. Assisted-by: Codex:gpt-5 * fix(upscale): validate scale and clean failed HTTP outputs Require an explicit scale bounded to 1-16, check backend results, and keep destination cleanup owned by the allocating endpoint. Cover validation, result failures, cleanup, and successful URL responses. Assisted-by: Codex:gpt-5 * feat(studio): add inline image upscale workflow Assisted-by: OpenAI:gpt-5 * test(studio): serve upscale capability fixtures * fix(stablediffusion): default ESRGAN tiling to full frame * feat(gallery): add RealESRGAN anime upscaler * fix(stablediffusion): allow full-frame ESRGAN tiles * fix(routes): select explicit upscale models Give image upscaling dedicated usecase-filtered middleware and keep ordinary image models out of the endpoint. Mark and verify the legacy Diffusers x4 gallery model as an explicit upscaler. Assisted-by: OpenAI:gpt-5 [shell] * refactor(upscale): keep settings in backend options * test(config): keep upscale options backend-specific * fix: keep upscale routing metadata out of backend options --------- Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-11 00:30:09 +02:00
mudler-agentandEttore Di Giacinto	08871b1a6a	feat(audio): add YuE2 music generation (#12617) ... * feat(audio): forward sound generation params Pass string-valued sound generation parameters from the public request to the audio backend. Typed fields retain precedence over colliding generic parameters, and source audio remains unavailable in the public schema. Assisted-by: nib:gpt-5.6-sol Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): add YuE2 music model Assisted-by: nib:gpt-5.6-sol Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs(audio): document YuE2 sound generation Assisted-by: nib:gpt-5.6-sol Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(audio): reserve source input parameter Assisted-by: nib:gpt-5.6-sol Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-11 00:26:10 +02:00
mudler-agentandEttore Di Giacinto	24a9feecbb	feat(api): add OpenAI Decisions endpoint (#12603) ... * feat: add OpenAI Decisions adapter with validated ordered answers Share SystemOne execution and preserve ordered, typed Decisions answers. Reject null probabilities and derive scores from the emitted distribution. Cover conversion and handler contracts using a valid generated PNG fixture. Assisted-by: nib:gpt-6-astra Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix: preserve Decisions choice types and empty score labels Keep canonical JSON values in model-visible choice criteria and represent probability label presence explicitly. Add conversion and serialization regressions. Assisted-by: nib:gpt-6-astra Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat: wire the OpenAI Decisions route Expose the adapter through the shared Decisions feature gate and usage middleware. Test real route registration with a fake backend, including unauthenticated access, user permissions, and exact usage accounting. Assisted-by: nib:gpt-6-astra Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: document OpenAI-compatible decisions and discovery Describe the shared local execution contract, typed questions, image limits, and backend accounting. Publish the Decisions route in Swagger and retain the existing SystemOne discovery tag. Use documented loose object placeholders for JSON unions that Swagger 2 cannot express. Assisted-by: nib:gpt-6-astra Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-10 08:29:31 +02:00
mudler-agent	cd604b5c5b	fix(distributed): bound, stop and cancel model loads with leases and worker operations (#12524) ... Fence model load jobs by generation, lease them on the database clock, bound the work on the worker with operations and a process-group watchdog, and add one stop path with a load-cancel API. See the pull request for the design, the rolling upgrade notes and the test evidence. Assisted-by: Claude Code:claude-sonnet-5-5 Signed-off-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-07 16:59:56 +02:00
localai-org-maint-botandEttore Di Giacinto	6b794651a4	feat(diarization): return sound events with include_sounds (#12544) ... * feat(diarization): return sound events with include_sounds A client that wants text, speakers, voice prints and sound events had to make a diarization call and a separate sound call. Add an include_sounds request field to /v1/audio/diarization that adds a sounds array of closed events {start, end, label, confidence}, in seconds. The parakeet-cpp backend runs a tagger-only scene stream over the clip, the same stream and thresholds the live path uses, so a clip gives the same events offline and live. A model with no sound_model companion, or a backend that does not report sound events, fails with 501 and the stable code include_sounds_unsupported instead of an empty list. The proto carries sounds_included so an empty list still means "nothing heard". The localai-proxy backend forwards the field. Swagger, docs and the e2e mock backend are updated. Assisted-by: Claude:claude-sonnet-5-5 [protoc swag go] * feat(gallery): add parakeet-cpp-multilingual-diarization-speakers-sounds Same as parakeet-cpp-multilingual-diarization-speakers (TDT 0.6B v3, Nemotron-3-Diarization, WeSpeaker) plus a CED-Tiny sound_model, so one model name serves /v1/audio/diarization with include_text, include_speaker_profiles and include_sounds. It declares the sound_classification usecase like the realtime scene entries. Assisted-by: Claude:claude-sonnet-5-5 --------- Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-07 14:10:45 +02:00
mudler-agentandEttore Di Giacinto	79a7631cc5	feat(parakeet-cpp): encoder fingerprint for speaker naming, VAD trim and word filter options, pin bump (#12491) ... * chore(parakeet-cpp): bump parakeet.cpp to 2de154c Brings in the speaker registry encoder fingerprint, the VAD segment trim and the opt-in word filter, a fix for a per-call thread count that stayed set on the process-wide backend after a Silero VAD pass, and bundle components loaded from memory. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * feat(parakeet-cpp): encoder fingerprint for speaker naming, vad_trim and guard_* options Speaker naming. A registered voice now carries the encoder that made it: the embedding family (voicedetect:<arch>:<name>:<dim>) and the sha256 of the weights. The backend reports the family of the loaded speaker model in Status, voice enrollment from speaker_profiles stores it as encoder_family (old entries load without it), and the registry sent to parakeet.cpp is built with parakeet_capi_speaker_registry_add_embedding_fp. The library then refuses a registry of another encoder family and the error names both families; another quantization of the same family only warns. A voice with only a weights hash gets the loaded family when the hashes are equal. Voices without a fingerprint (registered from audio: libvoicedetect cannot report one) keep the file-name rule and are used with a warning. The library cannot mix them with fingerprinted voices in one registry, so a request that has any uses the old registry for all. speaker_strict:true drops them instead. A library without the symbols behaves as before. Transcription. vad_trim (seconds, 0 keeps the whole cuts) goes through the VAD options JSON, so it reaches /v1/vad and the segmenter. The guard_* options guard_min_local_conf, guard_local_radius and guard_drop_punct_only turn on the word filter through parakeet_capi_transcribe_path_json_with, or through the segmenter with vad:true. They are off by default, bad values fail the load, and a library without the symbol fails it with a clear message. The dropped word count is logged at debug level. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] --------- Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-05 15:57:16 +02:00
mudler-agentandEttore Di Giacinto	99043b442c	feat(router): route with native decision models (#12449) ... * fix(schema): preserve SystemOne image inputs Assisted-by: OpenAI * test(schema): follow Ginkgo conventions for decision inputs Assisted-by: OpenAI * feat(llama-cpp): dispatch native decisions through Score Upgrade the stock dependency and reconcile Score/TTS patches. Reuse native decision parsing, tasks, formatting and response-reader cleanup; preserve ordinary scoring admission and guard older dependencies. Assisted-by: OpenAI * refactor(systemone): share request and model validation Assisted-by: OpenAI:gpt-5 * fix(systemone): preserve HTTP wire-byte validation limit Keep structural validation separate from the serialized internal request bound so HTML escaping cannot reject valid HTTP payloads. Assisted-by: OpenAI:gpt-5 * feat(systemone): bound images and account native decisions Preserve public wire limits independently from router serialization. Reject unsupported NER images, map native request/capability errors, and stamp explicit usage once. Advertise decisions for stock llama-cpp. Assisted-by: OpenAI * fix(systemone): record usage on registered native route Exercise real registration and billing with a mock native backend. Reject empty native responses, malformed image URLs, trailing JSON, and wire overflow including whitespace. Assisted-by: OpenAI * feat(router): add lazy native decision transport Bind named models through internal ModelSystemOne calls with shared validation and bounded abandoned operations. Remove request and echoed-error contents from decision traces. Assisted-by: OpenAI:gpt-5 * feat(router): classify overlapping policies with native decisions Ask independent noul questions, validate probabilities and preserve first-superset routing. Wire the central factory with config-sensitive invalidation and cancellation-safe resolution. Document native framing and bounded operation limits. Assisted-by: OpenAI:gpt-5 * feat(gallery): add pinned Julia-1 native decision model Add a separate text-only llama-cpp Q8 entry with pinned Apache-2.0 source provenance and checksum. Installed using the gallery installer and exercised choice, score and noul on CPU. Assisted-by: OpenAI * test(router): verify native decisions through central factory Add an opt-in real-model Ginkgo integration covering the native Go loader and C++ transport, token usage, independent overlapping labels, and candidate selection. Document owned-server execution and the intentionally non-quality threshold. Assisted-by: Codex:gpt-5 * fix(llama-cpp): align upstream pin and preserve decision signatures Advance to bed0a856 without losing the automated upstream bump. Detect full-request fill_task support at compile time and forward every question for Nimble framing while retaining the earlier native signature. Preserve reconciled SCORE/TTS patches; add standalone compatibility coverage. Assisted-by: Codex:gpt-5 * feat(gallery): add native decision family defaults Pin Laya, Kev-4B, lev, OpenJev and Nimble artifacts. Verify Laya/Kev/lev gallery installs and CPU contracts on both native pins; clearly mark OpenJev/Nimble runtime validation pending and their noncommercial licenses. Assisted-by: OpenAI * docs(decisions): clarify integrated Nimble prerequisite Record the exact combined backend pin while retaining pending OpenJev and Nimble installation/runtime validation status. Assisted-by: Codex:gpt-5 * fix(gallery): indent native decision model sequences Match repository yamllint indentation for Laya, Kev, lev and OpenJev list fields. Parsed gallery data is unchanged; reproduce CI gallery lint failure before the whitespace-only fix and pass the same command afterward. Assisted-by: Codex:gpt-5 * docs(decisions): record OpenJev and Nimble CPU validation Record gallery installation, checksum/metadata verification and multiquestion native smoke results on bed0a856. Retain noncommercial and text-only limitations without accuracy or deterministic-output claims. Assisted-by: OpenAI * fix(ui): expose native Decisions router classifiers Select classifier models using metadata-driven capability routing, retain tuned thresholds, and validate native decision selections before saving. Cover both native backends and create/save/reopen in the real React editor. Assisted-by: Codex:gpt-5 * fix(router): exclude aliases from native decision discovery Check the originally named config before advertising native Decisions eligibility. Retain target capability inheritance for ordinary generation aliases. Exercise the actual capabilities endpoint with native models on both backends, aliases, and disabled models. Assisted-by: Codex:gpt-5 * feat(systemone): share bounded multimodal input validation Preserve text wire limits while admitting bounded PNG/JPEG decision input. Share collection and header validation across internal and public callers and keep the native runner response budget independent. Assisted-by: OpenAI:API-assistant * fix(systemone): bound admission lifetimes and validate complete images Retain shared admission leases through actual work completion, including abandoned internal operations. Decode bounded image pixels, cap public native responses before usage stamping, and preserve oversized malformed text status precedence. Assisted-by: OpenAI:API-assistant * fix(router): classify images before media fetching Preserve ordered structured probes for native decisions. Defer OpenAI media preparation until routing selects the served model, so rejected decision URLs cannot trigger downloads before shared validation. Guard direct image collection with context-aware shared admission. Keep text classifiers and embedding caches from discarding image input. Retain fail-closed classifier configuration and runtime fallback policy. Add middleware, typed-content, admission, cancellation and cache tests. Assisted-by: OpenAI:API-assistant * fix(router): bound extraction before serialization Check probe budgets before copying text or marshaling message state. Count JSON escaping so oversized internal inputs fail before allocation. Preserve typed Anthropic blocks through selected-model conversion and fallback. Keep retry coverage in Ginkgo without global test registration. Assisted-by: OpenAI * fix(router): bound supported probe serialization Arbitrary structs can bypass the probe budget through pointer marshalers, string tags, and promoted fields. Accept concrete chat schema types and plain JSON values instead of emulating arbitrary struct serialization. Budget escaped direct prompts before marshaling so raw length cannot hide serialized expansion. Preserve runtime fallback and reject oversized input before invoking the decision runner. Add Ginkgo allocation, boundary, and marshaler invocation regressions. Six-package tests, three-package race tests, and full-T2 delta lint pass. Assisted-by: OpenAI:GPT-5 golangci-lint * feat(decisions): enable bounded OpenJev images Validate native decision images before permissive media parsing and pixel allocation. Require both decision image support and a vision projector; missing or audio-only projectors cannot silently become text decisions. Pin the OpenJev Q8 projector and document its license and disk footprint. Add native safety tests, canonical limit parity, gallery and load-option checks, and a reproducible CPU direct-RPC contrasting-image smoke. Assisted-by: OpenAI:GPT-5 * fix(decisions): reject incomplete image streams stb accepts corrupt PNG Adler checksums and truncated JPEG scans. Use bounded zlib validation and strict libjpeg decoding before parsing. Keep dimension and aggregate pixel checks ahead of decoder allocations. Wire decoder dependencies into native builds and runtime packaging. Add regressions for appended EOI and embedded marker bypasses. Assisted-by: OpenAI:GPT-5 * fix(ci): gate native decision image validation Run the decoder security tests outside the stdlib-only native suite. Fetch vendor headers at the backend pin and provision decoder dependencies. Gate Go limit parity and production CMake wiring without model downloads. Assisted-by: OpenAI:GPT-5 * test(decisions): cover multimodal public API paths Exercise shared image contracts through the registered HTTP routes and external mock backend. Add opt-in cached gallery installation and real OpenJev image decisions through SystemOne and both routing APIs. Assisted-by: Codex:gpt-5 * test(decisions): assert isolation and cache bypass Observe external RPC calls and compare complete classifier history. Winner-only and cache-miss checks could hide dropped history or cache use. Give real inference its own application and model directory so shared backend mappings and loaded processes cannot affect mixed suite order. Assisted-by: OpenAI:ChatGPT * test(decisions): isolate fixture globals Disable optional global services in the isolated HTTP fixture and register cleanup before setup assertions. Verify meter provider identity survives fixture creation and destruction. Snapshot observed usage before assertions so failures cannot retain the mutex. Require a successful usage stamp before checking error responses. Assisted-by: Codex:gpt-5 golangci-lint * fix(application): honor optional telemetry controls Skip failover gauge registration when metrics are disabled. Register against the application meter rather than looking up the global provider. Allow embedders to retain the bounded routing log without billing stats. Keep the existing default when stats are disabled. The isolated HTTP fixture uses this option without losing its native router assertions. Assisted-by: Codex:gpt-5 golangci-lint --------- Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-04 09:34:21 +02:00
mudler-agentandEttore Di Giacinto	9eb5a9e61d	feat(audio): remember speakers from diarization (#12414) ... * feat(schema): validate portable speaker profiles Add the versioned profile schema for explicit speaker enrollment. Validate compatibility against separately supplied loaded-encoder metadata. Reject unusable speakers, invalid vectors, and inconsistent clean spans. This slice does not change HTTP routes, backend integration, or the UI. Assisted-by: OpenAI:unknown Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(parakeet): export profiles with transcripts Export opt-in speaker profiles and trusted encoder metadata. Replay registrations by ID so duplicate display names keep independent vectors. Use one profile-capable diarization for slots, names, and clean spans. Assign timestamped ASR words to those slots without a second diarization. Preserve legacy opt-out and no-ASR behavior, and propagate failures. Assisted-by: OpenAI:unknown Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(audio): enroll portable speaker profiles Gate profile exports with voice-recognition permission and validate registration against metadata from the loaded encoder. Preserve audio enrollment and independent registrations with duplicate display names. Exclude diarization and registration exchanges before API trace capture so persisted traces cannot retain profile vectors or JSON audio. Defer candidate dimensions to trusted loaded metadata. Sort candidates by registration ID so incompatible profiles cannot suppress legacy voices through registry iteration order. Keep portable identity checks closed when trusted metadata is unavailable. Test persisted traces, explicit slot zero, and selection through offline and live transport. Document privacy and the ephemeral registry lifecycle. Assisted-by: OpenAI:unknown Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): remember speakers from diarization Add a Studio page for diarization and opt-in speaker profiles. Preview clean intervals from the original recording before explicit registration. Join profiles by raw speaker labels, preserve duplicate names, and relabel turns only after a successful save. Discard stale results when the model or recording changes. Share registration metadata with voice management without storing vectors or recordings from this flow. Document permissions and the global, ephemeral registry. Cover enrollment, permissions, previews, and asynchronous races with mocked Playwright tests. Assisted-by: OpenAI:unknown Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: clarify HTTP speaker enrollment support Replace the stale enrollment limitation with the current HTTP workflow. Distinguish native transport from explicit registration and link its docs. Assisted-by: OpenAI:unknown Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore(parakeet): pin merged speaker profile support Use the merged commit from mudler/parakeet.cpp#80. Its tree matches the previously accepted native pin. Assisted-by: OpenAI:unknown Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: add diarization enrollment setup example Connect the existing gallery modes to the speaker enrollment workflow. Show installation, private profile export, explicit raw-slot registration, and later recognition without another export. Assisted-by: OpenAI:unknown Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs(blog): explain diarization speaker profiles Put the diarization walkthrough on the LocalAI website in the feature PR. Cover the three gallery modes, explicit enrollment, and privacy limits. Link setup instructions and keep availability conditional on feature support. Assisted-by: OpenAI:unknown Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs(blog): focus diarization on everyday use Explain what users can do with recordings before the setup steps. Replace the technical walkthrough with a short Studio guide and link readers to the existing reference for model names and developer use. Assisted-by: OpenAI:unknown Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs(blog): lead with speaker capabilities Present speaker recognition through everyday uses and a short UI flow. Keep technical reference details in the existing documentation. Assisted-by: OpenAI:unknown Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(diarization): satisfy Go lint checks Avoid copying protobuf message state when extending backend status, check the multipart reader close result, and document the focused testing.T lint exemptions. Assisted-by: nib:gpt-5.6-sol Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-02 08:14:00 +02:00
localai-org-maint-botandEttore Di Giacinto	2ae6cae70d	feat(parakeet-cpp): name speakers from the shared voice registry (#12382) ... * feat(voice): list registered voices and record which encoder made them The voice registry could register, identify and forget but not list, and it did not remember which speaker encoder produced an embedding. Add Metadata.Model and Registry.List, answered from the index the store registry already keeps for Forget. Needed so a backend can be given the registered voices that match its own speaker encoder. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * feat(voice): store the encoder model with a registered voice Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * feat(voice): pick the registered voices that match a speaker model Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * feat(proto): carry known voices and speaker names on diarize and live messages Assisted-by: Claude:claude-haiku-4-5 [Claude Code] * feat(diarization): name speakers from the voice registry When a diarization model has a speaker_model option, the endpoint sends the registered voices made by that encoder to the backend. The backend's name and name_score come back as extra fields next to the normalized SPEAKER_NN speaker, and the speakers summary carries the first name seen for each speaker. RTTM output and results without names are unchanged. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * feat(live): pass registered voices to a live session and surface speaker names Live sessions now send the registered voices that match the model's speaker_model to the backend, and each speaker segment carries the name the backend matched. The realtime segment event gains an optional speaker_name field. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * feat(parakeet-cpp): load a speaker model and build per-request voice registries Adds the speaker bindings (ABI v9 and v10, probed separately), the speaker_model, speaker_threshold and speaker_margin options, and a per-request registry builder over the known voices. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * feat(parakeet-cpp): name the speakers in Diarize from the known voices Diarize builds a per-request speaker registry from the known voices when a speaker model is loaded, calls the named C functions, and puts each slot's registered name and score on the segments. The registry is freed on every path. A library without ABI 10 reports Unimplemented instead of dropping the names. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * feat(parakeet-cpp): name speakers in the live scene stream The live scene stream now begins with a known-voice registry when a speaker model is loaded and the live config carries voices, and each closed speaker segment takes its slot's current name from the feed's names map. A segment that closes before its slot is identified has an empty name. The registry is freed after the stream, on every path. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * feat(gallery): speaker naming entries and docs for parakeet-cpp Add three gallery entries that load the WeSpeaker ResNet34 speaker model next to the diarization or realtime scene models, and document speaker names in the voice recognition, diarization, audio to text and realtime pages. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * fix(parakeet-cpp): skip an unusable registered voice instead of failing the request A registered voice with the wrong embedding size, or one the C side refused, failed the whole diarization request, so one legacy voice broke the model for every user. Skip such voices with a warning that does not carry the voice name, and take the plain path when none is left. Also map an exact 0 speaker threshold or margin to a tiny positive value, since the C side reads 0 as "use the default", and fix a stale comment about which contexts Free() walks. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * fix(diarization): warn once per model about voices from another encoder; document the privacy limit The different-encoder warning fired on every request. Log it once per feature and speaker model, then at debug level. Document that the global voice registry lets any caller of a speaker_model model learn matching names, and that skipped wrong-sized voices are logged. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] * chore(parakeet-cpp): bump parakeet.cpp to 8c8cec0 (C-API v10) and check speaker naming against the real library The pin moves from 623a968 to 8c8cec0, which brings in everything merged in parakeet.cpp since: the voice identification change (C-API v9, #78) and raw-embedding enroll plus diarize-only speaker naming (C-API v10, #79). New real-library specs (gated on PARAKEET_BACKEND_TEST_SPEAKER_MODEL, _DIAR_MODEL, _WAV and, for the live path, _STREAM_MODEL) name the two speakers of two_speakers.wav from a committed pair of WeSpeaker embeddings, with the voices passed in reversed order. They also check that the float32 threshold reaches C through purego. The shared test loader now registers the v9/v10 and scene symbols as main.go does. The rebase onto origin/master had no conflicts. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] --------- Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-10-01 08:25:52 +02:00
3 people	2fa36e7147	feat(parakeet-cpp): speaker diarization, sound detection and live scene events (#12335) ... * feat(parakeet-cpp): load diarization and CED models and companions Repin PARAKEET_VERSION to parakeet.cpp PR #75's head, which adds parakeet_capi_model_kind (ABI v8). Bind the new diarization, sound event and combined scene stream C symbols through the same purego.Dlsym probe pattern already used for the batched JSON entry point, so the backend still loads against an older libparakeet.so. Load now classifies the loaded GGUF by role (ASR, diarization or sound) via parakeet_capi_model_kind and can load up to two companion models from Options[] (asr_model:, diarization_model:, sound_model:, paths resolved against opts.ModelPath), verifying each companion's kind and freeing every context opened so far on any failure. Free releases the primary and every companion. AudioTranscription now names the loaded role when it is not ASR instead of a generic model not loaded error. The dynamic batcher starts only when an ASR context ends up loaded, primary or companion. This is groundwork only: the Diarize and SoundDetection RPCs and the live scene stream that actually use these new roles land in later commits. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(parakeet-cpp): reset role fields on a failed companion load loadRoles' freeLoaded only released the C contexts it had opened; it left ctxPtr/diarCtx/tagCtx and companions pointing at those now-freed contexts, so a later Free() on the same instance would double-free. Zero all four alongside the CppFree calls. Also route AudioTranscriptionStream and AudioTranscriptionLive through notASRError when ctxPtr is unset but a diarization or sound model is loaded, matching AudioTranscription: both used to return the generic model-not-loaded error instead of naming the loaded role. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(parakeet-cpp): add speaker diarization Implement the Diarize RPC for the parakeet-cpp Go backend, wired to Nemotron-3-Diarization through libparakeet.so's diarization C-API. Plain diarization uses parakeet_capi_diarize_pcm; when include_text is set and an ASR companion is loaded, parakeet_capi_transcribe_and_ diarize_json fills each segment's text instead. Speaker labels are the decimal index, or "unknown" for -1 (no diarized speaker overlaps). min_duration_off merges same-speaker segments across a short gap before min_duration_on drops the segments still too short, then ids are renumbered. num_speakers/min_speakers/max_speakers/clustering_ threshold have no Sortformer equivalent and are logged at debug instead of rejected. Verified against the real Nemotron-3-Diarization + parakeet-tdt_ctc- 110m checkpoints on the two_speakers.wav fixture: correct A-B-A-B speaker segmentation and matching speaker-attributed transcripts. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(parakeet-cpp): add sound event detection Wire the SoundDetection RPC to the CED tagger context (p.tagCtx) loaded by Task 1's role classification. It runs the whole clip through a one-shot parakeet_capi_sound_stream_* session (window 10s, hop 10s, top_k set to the tagger's class count so every drained window carries a full score list), averages each class's score across the drained windows, sorts descending, then applies the request's threshold and top_k (0 keeps every class). No tagCtx returns FailedPrecondition; a libparakeet.so missing the sound_stream symbols returns Unimplemented. Every C call runs under engineMu, and the stream is always freed, even when a feed or drain call fails partway through. Verified against a real ced-tiny-q8_0.gguf on the rooster.wav demo clip: "Chicken, rooster" tops the list at score 0.91. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(parakeet-cpp): cancel sound detection mid-feed, shrink the lock SoundDetection now checks ctx before each 10 s feed slice (mirroring driver.go's feedSlices) and returns Canceled if the caller gave up, so a long clip can be interrupted instead of feeding to completion regardless. The stream is still freed on every path, cancellation included. Also narrow engineMu to the C calls: the drained JSON document is now decoded after the lock is released, splitting soundStreamScores into a locked soundStreamDrain (opts, begin, feed, drain, free) and an unlocked json.Unmarshal. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(parakeet-cpp): stream speaker and sound events during live transcription Add two additive proto fields, LiveSpeakerSegment and LiveSoundEvent, repeated on TranscriptLiveResponse. When a diarization or sound companion model is loaded, AudioTranscriptionLive now runs a no-ASR scene stream (parakeet_capi_scene_stream_begin) beside the ASR streaming session, feeding it the same PCM slices and forwarding any closed speaker or sound events alongside the matching ASR delta, or on their own when a slice has no ASR output. The scene stream is freed and reopened on a mid-stream Config reset, flushed with is_last before the closing FinalResult, and degrades gracefully (a warning, not an error) when begin or a later feed call fails, so live transcription keeps working ASR-only. Existing live behavior is unchanged when no companion is configured, and no scene C call is made in that case. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(parakeet-cpp): keep scene events off the ASR critical path in live Emit each slice's ASR result right after the ASR feed, before the scene feed for that slice runs, so a companion diarization/sound model never adds scene compute latency in front of the delta or <EOU> that drives realtime turn detection. Closed speakers/sounds go out afterward as their own response, so a slice with both now produces two responses, ASR first. The live feed log line now reports ASR and scene wall time separately. Re-check the diarization/sound contexts a scene stream was begun with against the live contexts before every feed, under the same lock: Free() can race between an ASR feed and the matching scene feed and free the model the stream borrows. A mismatch now returns without touching the C side. Freeing the stream itself stays unconditional; the scene stream's destructor only releases its own buffers and never touches the borrowed contexts. Also recover a panicking stub inside the live test goroutine instead of crashing the test binary, and reset the live decode-lag tracker on a mid-stream config reset, matching what its own comment already promised. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(realtime): surface live speaker and sound events Carry the backend's closed speaker segments and sound events (TranscriptLiveResponse fields 7/8) through LiveTranscriptionEvent as LiveSpeakerSegment/LiveSoundEvent (nanoseconds mapped to seconds), and forward them from the semantic_vad live path. Each speaker segment emits conversation.item.input_audio_transcription.segment with speaker, start, end and empty text under the turn's item id. Each sound event emits conversation.item.sound_detection with one tag (label, score = peak, index) and the event's new optional start/end seconds fields, omitted when unset so the existing unary/windowed sound-detection path is unaffected. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(realtime): keep start/end on a zero-second transcription segment ConversationItemInputAudioTranscriptionSegmentEvent.Start/End used omitempty, so a speaker segment starting at 0.0s dropped its "start" key. Nothing emitted this event before the live scene-event path, so drop omitempty: the segment always carries real times. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore(gallery): add parakeet-cpp diarization, CED and realtime scene models Add gallery entries for the new parakeet-cpp capabilities: standalone Nemotron-3-Diarization, the same paired with the Parakeet TDT+CTC 110M ASR model for speaker-attributed text, CED-Tiny and CED-Base sound classifiers, and a realtime scene bundle combining the streaming EOU ASR model with diarization and sound companions. SHA256 taken from the Hub API; licenses from each model card (openmdw-1.1 for Nemotron-3-Diarization, apache-2.0 for CED, cc-by-4.0 for the Parakeet ASR models). Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: document parakeet-cpp diarization, sound detection and live scene events Cover the new parakeet-cpp capabilities across the feature pages: Nemotron-3-Diarization as a diarization backend (with and without speaker text, the ignored speaker-count hints, the Sortformer voice-like-sound quirk), CED as a sound classification backend, the asr_model/diarization_model/sound_model/diarization_latency companion options, and the realtime live speaker/sound events (event shapes, the speech-turn-only limitation, and using this or pipeline.sound_detection but not both). Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(gallery): correct the realtime-scene license and wording nits parakeet-cpp-realtime-scene mistakenly copied cc-by-4.0 from the existing realtime_eou_120m-v1 entry; the model card lists the NVIDIA open model license instead. Switch to the gallery's usual spelling for that license and keep the diarization/CED licenses called out in the description. Also: audio-diarization.md now says getting per-segment text needs both an asr_model companion and include_text=true on the request, and audio-to-text.md's option table reads "Use on" (a pairing the loader does not enforce) instead of "Allowed on". Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(parakeet-cpp): reject a companion role that duplicates the primary's loadRoles let a companion option (asr_model:/diarization_model:/ sound_model:) assign into a role field the primary already occupied, for example asr_model: on an already-ASR primary. The companion's context silently overwrote ctxPtr/diarCtx/tagCtx, and Free() only walks those three fields, so the original primary context was never freed again. Reject a companion whose role the primary already holds before its GGUF is even loaded, freeing everything loadRoles opened so far, the same way a wrong-kind companion is already rejected. Also warn, rather than silently fall through, when parakeet_capi_model_kind reports PARAKEET_MODEL_KIND_NONE for a successfully loaded primary; the primary is still treated as ASR, matching today's behavior. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(parakeet-cpp): cap live scene sound score retention sceneBegin started the live diarization/sound companion stream with the C API's default sound options, whose top_k keeps 5 scores per window forever until drained. The live scene path never drains sound scores (only the offline SoundDetection RPC does, with its own fresh stream), so this window queue on the C side grew for the whole session's lifetime. Set opts.Sound.TopK = 0 before starting the scene stream: this disables score retention while leaving sound event detection (onset/ offset), which the live path actually consumes, unaffected. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(parakeet-cpp): merge diarization segments per speaker, harden Diarize mergeCloseSegments only compared neighbors in the single start-sorted segment list, so two same-speaker segments never merged once another speaker's turn fell between them (A, B, A): the short B segment broke the adjacency the merge relied on. Group segments by speaker first, merge within each speaker's own start-ordered run, then re-sort the result by start so interleaved speakers come back out in timeline order. Also harden Diarize's entry points the same way streamFeedDoc/ sceneFeed already are: diarizeCall re-checks p.diarCtx (and, on the include_text path, p.ctxPtr) under engineMu right before the C call, so a Free() racing between Diarize's own checks and the lock can no longer reach the C side with a freed context. When the include_text call returns NULL, last_error is now read from both contexts and whichever came back non-empty is reported, since either side of the pairing can be the one that failed. A WAV decode failure is reported as InvalidArgument instead of an unwrapped/untyped error. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(parakeet-cpp): harden SoundDetection's engine checks soundStreamDrain ran every C call under engineMu but never re-checked p.tagCtx there, so a Free() racing between SoundDetection's own tagCtx==0 check and this lock could still reach the C side with a freed context. Re-check p.tagCtx under the lock and return ModelNotLoaded when it was cleared, mirroring diarizeCall's own re-check. A WAV decode failure is now reported as InvalidArgument instead of an unwrapped/untyped error. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(parakeet-cpp): cover a mid-session scene feed failure feedSlicesScene already degrades gracefully when a scene feed call fails mid-session: it frees the broken stream and carries the ASR-only session forward. Add a spec covering that path end to end: the scene stream is freed exactly once, later audio slices still produce ASR responses, and no speaker/sound events appear before or after the failure. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: fix the parakeet-cpp companion role table and realtime scene docs audio-to-text.md's companion option table read "Use on" with a note that the loader did not enforce the pairing; it now rejects a companion whose role duplicates the primary's, so restore the "Allowed on" wording and describe the real enforcement. openai-realtime.md's live speaker/sound section claimed a mid-stream session.update resets the companion stream and that it flushes on session close; neither happens, since the realtime core opens one live stream (and so one scene stream) per speech turn and closes it at that turn's commit, with no mid-stream Config in between. Document that lifecycle instead, state precisely that start/end are seconds from the start of the turn's own audio, and note that the diarization model starts a fresh session every turn, so a speaker index is only meaningful within one turn. The example sound tag ("Rooster", index 17) did not match any real CED label; index 17 in ced-tiny-q8_0.gguf is "Baby laughter". Replaced with "Chicken, rooster" at its real index, 99. Assisted-by: Claude:claude-sonnet-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(parakeet-cpp): use CED's real index for Chicken, rooster The scene feed comment and the live test's canned document gave "Chicken, rooster" index 365. In CED's AudioSet label list it is 99, which is also what the realtime docs show. Assisted-by: Claude:claude-opus-5-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(realtime): call the test event accessor The scene-event tests range over a method instead of its returned slice. Call the synchronized accessor so the OpenAI test package compiles. Assisted-by: Codex:gpt-6 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore(parakeet-cpp): pin parakeet.cpp master with sound events mudler/parakeet.cpp#75 (sound events, scene stream, model kinds) and #74 (the missing <algorithm> include that broke the image builds) are on master now. Pin 6dea76a instead of the #75 PR head, and update the header comment the bump bot reads. Assisted-by: Claude:claude-opus-5-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore(parakeet-cpp): pin parakeet.cpp with ced.cpp on main parakeet.cpp #76 moved its ced.cpp submodule from the head of localai-org/ced.cpp#3 (a branch-only commit) to ced.cpp main, where #3 landed with an identical tree. Pin 623a968 so the image builds no longer depend on that branch. Assisted-by: Claude:claude-opus-5-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(transcription): carry speaker labels on words and streamed segments A diarizing backend could label transcript segments, but two paths dropped the label: TranscriptWord had no speaker field, so live transcription words and word-level timestamps could not carry one, and the stream=true transcript.text.done event left the speaker out of its segments. TranscriptWord gains an optional speaker (proto field 4, additive). It flows through the live event and result mapping, the JSON word output of the endpoint and the CLI, and transcript.text.done now includes a segment's speaker when there is one. Empty labels are omitted, so responses without diarization are unchanged. Assisted-by: Claude:claude-opus-5-5 [Claude Code] (cherry picked from commit 2f0049f979) Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(importers): detect the parakeet.cpp diarization GGUF The Nemotron-3-Diarization GGUFs are published in mudler/parakeet-cpp-gguf as nemotron-3-diarization-<quant>.gguf. The parakeet-cpp importer did not recognise that name, so a direct `local-ai models import` of the file fell through to another importer. A direct URL to the file now imports with the diarization usecase. A repo import still picks ASR weights when the repo also ships the diarization model, and falls back to the diarization weights only when there are no others. Ported from #12323. Assisted-by: Claude:claude-opus-5-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(config): advertise diarization and sound detection for parakeet-cpp The capability table listed parakeet-cpp as transcription only, though the backend now answers Diarize (Nemotron-3-Diarization) and SoundDetection (CED) depending on the model kind it loads. Assisted-by: Claude:claude-opus-5-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(parakeet-cpp): label transcript segments with the diarization companion A diarization_model companion only fed live speaker events and Diarize; /v1/audio/transcriptions ignored it. With the companion attached and diarize=true (the OpenAI endpoint's default), unary transcription now labels each segment with its speaker and splits segments at speaker turns; with word timestamps each word carries its speaker. The stream=true final result labels each utterance with the speaker who said most of it. Both use the checkpoint's own diarization over the whole clip, as NeMo's diarize() does. Words take the speaker whose segments overlap them most, or the nearest segment within 0.5 s, the same rule as parakeet.cpp's speaker-attributed ASR. Docs: the diarization_model row and a paragraph on transcript speakers; Nemotron-3-Diarization handles up to 8 speakers. Ported from #12323. Assisted-by: Claude:claude-opus-5-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(realtime): speaker segments from committed-turn transcription Speaker events reached a realtime session only from the live semantic_vad path, which needs a cache-aware streaming transcription model. Committed-turn transcription (server_vad, or any offline model) always asked the backend for diarize=false and dropped the segments' speakers. pipeline.diarization (off by default) asks the transcription model for speaker labels on each committed turn and emits every labelled segment as a conversation.item.input_audio_transcription.segment event, with its text, before the turn's completed event. It is opt-in because some backends fail a diarization request they cannot serve. Assisted-by: Claude:claude-opus-5-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): add parakeet-cpp-realtime-scene-tdt parakeet-cpp-realtime-scene pairs the streaming EOU model with the diarization and CED companions; its speaker and sound events need a cache-aware streaming model. This entry does the same with Parakeet TDT 0.6B v3 (multilingual, offline) for realtime under server_vad: set it as both transcription and sound_detection and turn on pipeline.diarization, and each committed turn gets speaker segments and sound tags from one parakeet-cpp backend. Files and sha256 match the Hub and are shared with the existing TDT v3, diarization and CED-Tiny entries. A real-model spec checks the combination on a clip with two speakers and a rooster: A-B-A-B speaker turns, and "Chicken, rooster" among the sound tags. The test loader now binds the sound entry points like main.go. Assisted-by: Claude:claude-opus-5-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): add CED-Base variants of the parakeet-cpp scene models parakeet-cpp-realtime-scene and parakeet-cpp-realtime-scene-tdt ship with CED-Tiny. The -base variants use CED-Base (86M), which tags sounds more confidently (on the rooster clip "Crowing" 0.65 against 0.49 for Tiny). Measured on CPU over a 37 s clip: the live diarization + sound stream runs at 0.125 of real time with CED-Base against 0.103 with CED-Tiny, because diarization dominates; sound detection per committed turn costs 0.031 against 0.005. The realtime docs list both and note that any CED size works as sound_model. Files and sha256 match the Hub and are shared with the existing parakeet-cpp-ced-base entry. The TDT variant passes the real-model scene spec with CED-Base (A-B-A-B speakers, "Chicken, rooster" found). Assisted-by: Claude:claude-opus-5-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(config): register pipeline.diarization in the config metadata TestAllFieldsHaveRegistryEntries fails on the branch because the new pipeline.diarization field has no registry entry. Add one so the model editor shows it as a toggle next to the sound detection options. Assisted-by: Claude:claude-sonnet-5-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com>	2026-09-29 23:58:17 +02:00
leilei3167	afebecc63d	fix(ollama): report on-disk size for /api/tags and /api/ps (#11989) ... Hardcoding size/size_vram as 0 made Ollama clients treat loaded models as free. Prefer ModelFileName+ModelPath Stat when available, and omit size_vram (and size) when the value is unknown instead of emitting literal zeros. Resolve each listed model by its stored ID so tagged variants use their own weights. Fixes #11969 Signed-off-by: lei_lei <imleilei123@gmail.com>	2026-09-28 00:11:41 +02:00
localai-org-maint-botandlocalai-org-maint-bot	f154bd990a	feat(system): report per-model DRM VRAM (#12026) ... * feat(system): report per-model DRM VRAM Expose optional resident device memory for local backend process trees. Deduplicate DRM clients and omit unsupported or incomplete readings. Document accounting limits and preserve a measured zero in JSON. Closes #11970. Assisted-by: Codex:gpt-6 * fix(system): document trusted procfs reads Scope G304 annotations to paths built from the fixed procfs root, integer process IDs, and kernel directory entries. These reads accept no user-controlled path components. Assisted-by: Codex:GPT-6 gosec --------- Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com>	2026-09-27 21:18:34 +02:00
localai-org-maint-botandlocalai-org-maint-bot	4e94c914c9	fix(swagger): describe backend metadata as an object (#12178) ... Swag cannot resolve json.RawMessage in OpenAIResponse and aborts the daily schema generation. Set its Swagger type without changing JSON encoding, and regenerate the checked-in specifications. Assisted-by: Codex:gpt-6 Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com>	2026-09-27 21:06:54 +02:00
3 people	21c5495a99	feat(vllm-cpp): add GLiNER2.5 NER via TokenClassify (#12140) ... * feat(vllm-cpp): add GLiNER2.5 NER via TokenClassify Wire the vllm-cpp backend to the C ABI NER surface (vllm_gliner_ner, ABI v27) so LocalAI can serve zero-shot named entity recognition through the existing TokenClassify gRPC method. backend.go: TokenClassify method on *VllmCpp calls vllm_gliner_ner with the text and labels, copies the C-owned entity array into protobuf TokenClassifyEntity messages, and frees the result. govllmcpp.go: cNerEntity and cNerResult Go POD mirrors matching the C structs; vllmGlinerNer and vllmNerResultFree purego bindings; abiVersion bumped 26 -> 27. options.go: ner_labels, ner_threshold, ner_max_width parsed from engine_args. pkg/grpc: ClassifyModel interface and TokenClassify server handler (follows the Embedding locking pattern). core/config: vllm-cpp backend declares MethodTokenClassify and UsecaseTokenClassify. docs/content/features/vllm-cpp.md: NER section documenting the engine_args keys and the host-forward contract. Assisted-by: MAKI:regolo/glm5.2 [maki] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(vllm-cpp): correct NER pointer lint directive Use the govet directive for the C-owned NER array, matching the other purego pointer conversions. The array remains valid until its deferred free; the misspelled directive caused CI to flag this conversion. Assisted-by: Codex:gpt-6 golangci-lint Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(vllm-cpp): add kev-compatible SystemOne API endpoints Add POST /v1/systemone, /v1/systemone/permute, and /v1/systemone/separate to LocalAI, mirroring the kev project's structured-extraction API. Each endpoint runs zero-shot NER over the rendered state text and builds kev-compatible answers for three question types: noul (binary entity presence), choice (pick one option), and score (pick one level). The TokenClassifyRequest proto gains a `repeated string labels` field so each question can supply its own labels at inference time, and TokenClassifier gains TokenClassifyWithLabels for per-call label selection. The vllm-cpp backend uses request labels when non-empty, falling back to configured ner_labels then the built-in defaults. Helpers (renderState, softmax, choiceConfidence, scoreConfidence, r2) are ported from kev/api.py and mirrored in vllm.cpp's api_server.cpp so both servers produce the same answer shape. Following-Agents-Protocol: true AI-Assisted: true Assisted-by: AGENT:regolo/glm5.2 [maki] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(vllm-cpp): suppress gosec G404 on seeded permutation RNG The SystemOne permute endpoint uses math/rand with a caller-supplied seed for reproducible option permutations, matching kev's random.seed. gosec flags this as G404 (weak RNG). Add #nosec with a comment naming the intent: this is reproducibility, not cryptography. Following-Agents-Protocol: true AI-Assisted: true Assisted-by: AGENT:regolo/glm5.2 [maki] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore(vllm-cpp): bump vllm.cpp pin to GLiNER2.5 merge commit Advance VLLM_CPP_VERSION from f3cd97e to 5058268d, the commit that landed GLiNER2.5 zero-shot NER support (PR #3224) in vllm.cpp. This brings the DeBERTa v2 encoder, GLiNER2 boundary head, C ABI NER functions, and server endpoints into the LocalAI vllm-cpp backend. The ABI version (27) and Go struct mirrors already match. Following-Agents-Protocol: true AI-Assisted: true Assisted-by: AGENT:regolo/glm5.2 [maki] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(vllm-cpp): use instruction text as NER label in SystemOne handler The SystemOne handler was passing question IDs as NER labels for noul questions and bare key names for choice questions, so the model never matched any entities. Port the label mapping from vllm.cpp's ParseSystemOneBody: - noul: use the rendered instructions field (with instr alias) as the NER label, not the question ID - choice: use optionText(name, desc) — "name: description" or "name" when the description is null/empty — not the bare key - score: already correct (rendered criteria text) - permute: shuffle indices and build parallel key/label arrays so the NER call uses the optionText labels while the response is keyed by the original option names Also add the instructions field to the SystemOneQuestion schema struct (accepted alongside the instr backward-compat alias). Verified end-to-end against the real GLiNER2.5 model: noul questions now find "Apple Inc. is" (organization, 0.999) and "Tim Cook is" (person, 0.852) where they previously returned zero entities. Following-Agents-Protocol: true AI-Assisted: true Assisted-by: AGENT:regolo/glm5.2 [maki] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com>	2026-09-23 12:31:28 +02:00
localai-org-maint-botandEttore Di Giacinto	c0b7e64973	feat(ui): show running models and host gauges on single-node installs (#12189) ... On a single-node install nothing in Operate listed the models loaded on this machine or let an admin stop one. The System page that did was retired in #11548, and its replacements (the Nodes workbench) only work in distributed mode. The Nodes page also mis-detected single-node mode: the cluster routes are not registered there, so /api/nodes answers 404, but only 503 was treated as "distributed off", which sent every single-node install to the empty worker-registration card. The rail hid the entry anyway. Nodes route on a single node becomes "This machine": - the Nodes page's VRAM / RAM / CPU / models-disk gauges, fed from this host by mapping /api/resources onto the worker heartbeat fields - a memory bar splitting host RAM by running model - a running-models table (backend, RSS, CPU share, uptime, PID) with search, sorting, logs and a confirmed Stop - the distributed setup behind an "Add machines" button The Operate overview gains a "Running now" preview (heaviest five, with Stop) on single node and a pointer to Nodes > Running models on a cluster. The rail shows "This machine" in Runtime with a running count. Backend, additive only: - /system: each loaded model carries a `process` block (pid, rss_bytes, memory_percent, cpu_percent, started_at). A sampler keeps one gopsutil handle per PID so CPU is the share since the previous poll rather than the lifetime average; it is omitted on the first reading. - /api/resources: host `cpu` and models-path `disk`, the same readings workers send in their heartbeat. Also fixes the fleet tables widening the page on phones: the headers' absolutely positioned sr-only labels escaped the scroll wrapper. Assisted-by: Claude:claude-opus-5 [Playwright] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-09-21 19:12:35 +02:00
Richard Palethorpe	495165266f	feat(kimodocpp): track usage through generic backend metadata (#12162) ... Report input tokens and frame-step output units in response metadata and record them through the existing usage accounting pipeline. Preserve the accounting rule and model-specific dimensions as JSON without extending the gRPC schema for each modality. Expose animation usage only under metadata.usage, validate counts before recording, and document the response contract and loaded-model location. Add coverage for transport, defaults, failures, persistence, and recording requests once with statistics enabled or disabled. Assisted-by: Codex:GPT-6 Signed-off-by: Richard Palethorpe <io@richiejp.com>	2026-09-20 19:39:06 +01:00
mudler-agentandEttore Di Giacinto	96ccb8e96c	fix: correct stale symbol names in doc comments (#12139) ... Fixes 7 doc comments whose names no longer match their function: - ToggleStateModelEndpoint (was ToggleModelEndpoint) - Messages.ToProto (was MessagesToProto) - NewUserCancellableContext (was newUserCancellableContext) - CalculateRMS16 (was calculateRMS16) - NewLocalAIMetricsService + Shutdown TODO (was copied setupOTelSDK) Supersedes #12084 and #12085 (DCO not signed by contributors). Assisted-by: MAKI:regolo/glm5.2 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-09-19 23:39:44 +02:00
lqp	dbee3db07f	feat(openai): add negative_prompt field to image generation request (#12031) ... Expose a negative_prompt string parameter on the image endpoints, matching Stable Diffusion WebUI / vLLM-Omni conventions. When both the negative_prompt parameter and a '|'-suffixed negative prompt in the main prompt are present, they are joined with a comma so callers can keep a global negative prompt in negative_prompt and add per-image negative tags after '|'. Assisted-by: Pi: DeepSeek V4 Pro Signed-off-by: Fedor Zuev <Fedor.Zuev@gmail.com>	2026-09-18 23:28:14 +02:00
Richard Palethorpe	2facfc0d88	feat: Add kimodo.cpp and 3D animation API/UI (#12095) ... * fix(vulkan): preserve host ICD discovery for packaged backends Add bundled Mesa manifests through VK_ADD_DRIVER_FILES instead of replacing the system driver list. Merge inherited and model-specific additive paths while preserving explicit operator overrides, with regression coverage. Assisted-by: Codex:gpt-5 golangci-lint Assisted-by: Codex:gpt-5.6-sol Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(3d): add Kimodo CPU and Vulkan animation backend Introduce a distinct animation capability and model-described 3D operations, with a typed /3d/animate API, RPC transport, distributed media staging, permissions, and tracing. Add a persistent kimodo.cpp adapter, skeleton GLB export, CPU/Vulkan packages, model and backend galleries, importer support, CI builds, and documentation. Adapt Studio inputs to each model and provide real-time skeleton playback, seeking, and history. Cover backend validation, packaging, API behavior, importer inventories, distributed staging, and Studio workflows. Validate real-model CPU/Vulkan generation and deploy the integration to the local QA instance. Assisted-by: Codex:gpt-5 golangci-lint Assisted-by: Codex:gpt-5.6-sol Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(kimodocpp): adopt monolithic encoders and resident inference Update upstream for resident weights, packed execution paths, and cached motion graphs. Default to all 32 text layers while retaining configurable streaming and legacy bundle support. Use monolithic Q8_0 encoders by default and offer all six published quantizations through the gallery and importer. Refresh pinned hashes, tests, and documentation; remove the obsolete thread patch and ensure cached source checkouts follow the upstream pin. Validated CPU and Vulkan generation, lower-bit streaming, gallery/importer suites, packaging, lint, and cold/warm Studio generation on localai-dev. Assisted-by: Codex:gpt-5 golangci-lint Assisted-by: Codex:gpt-5.6-sol Signed-off-by: Richard Palethorpe <io@richiejp.com> --------- Signed-off-by: Richard Palethorpe <io@richiejp.com>	2026-09-18 06:12:03 +01:00
localai-org-maint-botandEttore Di Giacinto	2d30d28f32	feat: expose context_size in /v1/models/capabilities (#12083) ... * fix(ui): restore node operation controls The node restructure hid backend logs and split related controls across inconsistent layouts. Restore contextual log actions and align the detail page with the fleet dashboard. Make multi-node selection clear and accessible. Assisted-by: Codex:gpt-5 Playwright ESLint * feat: expose context_size in /v1/models/capabilities The /v1/models/capabilities endpoint already enriches the OpenAI models listing with capabilities and modalities. Add context_size to that enrichment so clients (e.g. nib) can size their context budget for auto-compaction and pruning without guessing. The value comes from backend.EffectiveContextSize, which resolves the configured context_size or falls back to DefaultContextSize (4096) — the same number the backend itself runs with. It is omitempty so a loose file with no config surfaces 0 (unknown) rather than a misleading default, and existing clients that don't know the field see no change. --------- Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-09-16 14:40:35 +00:00
Abdullah Mansour | عبد الله منصور	acc7284526	fix(openresponses): support Codex WebSocket warm-up (#11608) ... * fix(openresponses): support Codex WebSocket warmup Assisted-by: ChatGPT:GPT-5.6-Sol golangci-lint Signed-off-by: Abdullah Mansour <abdullahmansour.marketing@gmail.com> * docs(openresponses): document WebSocket responses Assisted-by: Codex:GPT-5.6-Sol gh Docker golangci-lint Signed-off-by: Abdullah Mansour <abdullahmansour.marketing@gmail.com> * fix(openresponses): harden WebSocket response lifecycle Ensure response ownership, continuation storage, error sequencing, and connection-local resource limits remain correct across HTTP and WebSocket transports. Assisted-by: Codex:GPT-5.6-Sol [gh] [Docker] [golangci-lint] Signed-off-by: Abdullah Mansour <abdullahmansour.marketing@gmail.com> --------- Signed-off-by: Abdullah Mansour <abdullahmansour.marketing@gmail.com>	2026-09-11 16:54:42 +02:00
localai-org-maint-botandEttore Di Giacinto	88d19567b8	feat(faces): replay saved face enrollments (#11908) ... Accept original embeddings and timestamps so clients can restore faces when the in-memory store restarts. Derive stable IDs from exact vectors to make registration retries preserve identity without duplicate entries. Assisted-by: Codex:GPT-6 golangci-lint Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-09-07 17:40:27 +02:00
localai-org-maint-botandlocalai-org-maint-bot	0761bd02c7	feat(chat): add end-to-end context compression (#11556) ... * feat(config): add context compression policy Define the opt-in model configuration contract before the chat middleware consumes it. Document each policy field so later request handling does not invent a second schema.\n\nRefs #9534\n\nAssisted-by: Codex:gpt-5 * fix(config): register compression fields The model editor metadata gate rejects new config fields without descriptions and suitable controls. Register the compression policy so operators can edit its six fields safely. Assisted-by: Codex:gpt-5 [monitoring-prs] * feat(chat): compress long contexts Long conversations currently fail once they reach the model context window. The opt-in policy now summarizes complete older turns before primary inference and preserves the newest tool chains. Both OpenAI and MCP chat routes share the same transformation. Usage metadata and metrics expose each compression event. Refs #9534 Assisted-by: Codex:gpt-5 --------- Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com>	2026-08-18 11:31:03 +00:00
Richard Palethorpe	d10374f849	feat(router): make KNN a first-class classifier with a persisted, curated corpus (#10652) ... * feat(router): make KNN a first-class classifier with a persisted, curated corpus Add `classifier: knn` — similarity-weighted voting over labelled example prompts. Unlike score/colbert it needs no classifier model: label knowledge lives in a corpus seeded and curated through the admin API, so routing decisions are deterministic, auditable, and grounded in graded experience rather than a model's opinion. Epistemic gate: corpus entries below knn.similarity_threshold cannot vote; when none clears it the classifier activates no labels and the router uses the fallback — a prompt unlike all labelled experience is treated as undecidable, not guessed. Decisions record nearest_similarity (also on fallback rows) so admins can see how far the nearest labelled experience was; the Routing tab explains out-of-corpus fallbacks and shows per-label corpus counts. Persistence: one JSONL file per router under <data path>/router-corpus (text, labels, vector, embedder fingerprint). The file is the source of truth; the local-store index is rebuilt from it at classifier build time and stays a pure in-memory index. Entries recorded under a different embedding model re-embed on load. Also corrects the docs' false claim that local-store collections persist — the embedding cache never survived restarts (and still doesn't); the corpus does. Corpus input is API-only by design (entries may contain example user content): POST /api/router/{name}/corpus seeds (labels validated against declared policies, embedded server-side, indexed immediately), GET .../corpus/stats inspects — label counts only, entry texts are never returned by any surface — DELETE .../corpus wipes. Admin-gated like the sibling router endpoints, and exposed as MCP tools (seed_router_corpus / get_router_corpus_stats / clear_router_corpus) in both the httpapi and inproc clients with coverage-test route mappings. Plumbing: VectorStore gains SearchK (top-K was hardcoded to 1); local-store gets InsertBatch/Delete as optional fast paths; RouterConfig gains a knn block (embedding_model, k, similarity_threshold, vote_threshold, store_name) with meta-registry fields; the classifier dropdown now offers knn and the previously-missing colbert; embedding_cache is ignored (with a warning) for knn — it IS an embedding-KNN lookup; the stale /api/instructions intelligent-routing entry is rewritten (it described a classifier that no longer exists); swagger regenerated. Tests: KNN vote/gate specs with hand-computed vote shares, corpus manager suite (restart reload without re-embedding, fingerprint re-embed, dedupe, hostile store names), middleware specs (corpus routing, gate fallback, config validation, cache-wrap refusal), corpus endpoint specs pinning the texts-never-returned contract, MCP catalog + route-mapping gates, and a Playwright spec for corpus stats and the out-of-corpus decision detail. Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(router): name consulted corpus neighbours in knn decisions Every knn decision (decision log rows and the /api/router/decide response) now carries neighbors: the K retrieved corpus entries by descending similarity - including ones below the epistemic gate, which is what makes fallback decisions diagnosable - each as {id, similarity, labels}. The id is the entry's content hash (first 8 bytes of the SHA-256 of its text, hex): stable across reseeds and re-embeds, and text-free, so an external platform that seeded the corpus can recompute text->id on its own copy and bucket decisions by corpus region (per- region reliability accounting) without corpus text ever leaving the server. A corrupt index payload surfaces as an id-less neighbour at a real similarity instead of disappearing. Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * refactor(router): deduplicate knn plumbing and cut corpus hot-path waste Post-review cleanup of the knn-first-class-router branch; no behaviour changes on the API surface. Reuse/altitude: - RouterKNNConfig.ResolvedStoreName is now the single source of the router-corpus-<name> default (was hand-derived in four files). - corpus.ResolveKNNRouter + corpus.Seed carry the shared model resolution and seed validation; the REST endpoints and the assistant MCP client are thin transport adapters over them, with sentinel errors mapped to HTTP statuses at the echo boundary. - middleware.NewClassifierDeps assembles the classifier dependency set once for all five entry points (OpenAI, Anthropic, realtime, decide, corpus) instead of five hand-copied literals. - router.AllClassifiers feeds both the status endpoint and the unknown-classifier error, ending the classifier-list drift. - Per-classifier requirements moved out of validateRouterPolicies into their buildClassifier arms; the knn arm owns its embedding_cache opt-out instead of a name-check in the shared wrap tail. - adminOnly replaces four inline copies of the admin gate in the middleware routes. - localVectorStore.Search delegates to SearchK (identical traces). Efficiency: - Manager.Add embeds outside the manager mutex and appends to the JSONL file (O(new) instead of O(corpus) rewrite); a torn tail from a crash mid-append is tolerated on read and repaired on next write. - Stats memoises per store keyed on the file's stat fingerprint and no longer takes the manager mutex, so the 5s status poll stops parsing vector-laden JSONL and stops blocking behind seeds. - KNN Classify decodes each neighbour payload once (was twice) and builds refs and votes in a single pass with one fallback return. - Corpus file writes fsync before rename/close. - The corpus manager is built eagerly in newApplication (sync.Once dropped); test helper dead branch removed. Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(router): bind knn corpus vectors to an embedder fingerprint and fail closed on mismatch Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * chore(mcp): align corpus tool prompts and the mutating-tool safety list Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(proto,backend): report embedding shape from the llama-cpp backend Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(embeddings): Go-side pooling — mean/last/decayed_mean with half-life Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(embeddings): accept chat messages[] and per-request pooling on /v1/embeddings Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * chore(middleware): name the failing fields when post-merge validation 400s An intermittent post-merge validation failure surfaced as an opaque 400 during integration (pooling scheme mismatch that no client had sent). Log the model, the request's pooling override, and the merged config's pooling fields at the failure point so the next occurrence identifies whether the request or the stored config carried the bad value. Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(embeddings): scheme override must not inherit the config's half-life A model config defaulting to decayed_mean pooling carries pooling_half_life_tokens; a request overriding the scheme to mean/last without its own half-life inherited that value, and post-merge validation rejected the pair the server itself had assembled. Zero the inherited half-life when the overridden scheme is not decayed_mean; a request that explicitly pairs a half-life with a non-decayed scheme still 400s. Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix embedding pooling validation and router bounds Declare backend embedding layouts and reject incompatible pooling modes. Reset local-store dimensions after a full clear, validate KNN thresholds, and add real backend and store integration coverage. Assisted-by: Codex:gpt-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * ci: run local-store integration tests Build and install the local-store backend in the Linux test job, then run the existing store integration suite so new specs are discovered automatically. Assisted-by: Codex:gpt-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> --------- Signed-off-by: Richard Palethorpe <io@richiejp.com>	2026-08-18 09:37:43 +02:00
3 people	88edd7fc7f	fix(distributed): run cold model loads as durable jobs instead of holding the advisory lock (#11514) ... * fix(advisorylock): set statement_timeout alongside lock_timeout WithLockCtx already overrides a deployment-wide lock_timeout on its dedicated connection so a blocking pg_advisory_lock() waits its turn instead of failing with 55P03. statement_timeout aborts that exact same statement independently, with SQLSTATE 57014, and was not overridden. Production roles commonly carry statement_timeout=60s. Any guarded section longer than that (a cold model load stages for tens of minutes) therefore killed every concurrent waiter: advisorylock: acquiring lock 9003261067483446873: ERROR: canceling statement due to statement timeout (SQLSTATE 57014) Derive it from the same context budget as lock_timeout, with a matching RESET so the pooled connection is returned clean. Assisted-by: Claude Opus 5 [claude-code] * feat(distributed): add ModelLoadJob, the durable cold-load record A cold load in distributed mode is a long-running background job, but it was modelled as a synchronous side effect of an inference request: the whole of it (backend install, multi-GB staging, checkpoint load) ran inside the per-model advisory lock. Loading a 35.7 GB GGUF held that lock for ~20 minutes, so every concurrent request for the same model blocked on pg_advisory_lock and died at the role's 60s statement_timeout. Introduce the row that lets the lock shrink to a decision. Exactly one ModelLoadJob may be active per tracking key; that uniqueness — not the lifetime of a lock — is what de-duplicates concurrent loaders across replicas. ClaimLoadJob does its read-then-write under the advisory lock and nothing else: no network, file or gRPC I/O inside the guarded section, so a claim costs milliseconds no matter how long the resulting load takes. LastProgress is a heartbeat rather than a byte counter. A checkpoint load legitimately moves zero bytes for many minutes, so a reaper keyed on byte movement would reclaim a healthy job mid-load; byte progress stays the concern of load_deadline.go. A job whose heartbeat stops for longer than the orphan window is reclaimable, so a replica killed mid-load cannot wedge a model permanently. Failed jobs keep their row for a short grace so an immediately-following request reports the real cause instead of silently starting a fresh load of a model that just failed. No caller yet — the router moves onto this in the next commit. Assisted-by: Claude Opus 5 [claude-code] * refactor(distributed): run cold loads as jobs, outside the advisory lock Route wrapped the entire cold load — node selection, backend install, multi-GB staging and the remote LoadModel — in the per-model advisory lock. The lock's job is to de-duplicate concurrent loaders, a decision that takes milliseconds; holding it for the tens of minutes the resulting work takes is what turned a dedup mechanism into a cluster-wide outage for that model. Split it into a claim and a run. The claim is the only thing left inside the lock. The run is a background job owned by the claiming replica and bounded by the same progress-extended deadline as before; every other request for that model — local or on another replica — attaches as a waiter and is served the moment the model is ready, with no duplicate load and no lock contention. Waiters share one broadcast rather than an ordered queue: they all want the identical outcome, so ordering them would add fairness machinery that changes no result. The local channel wakes same-replica waiters instantly and a 2s DB poll is the authority, because a waiter on another replica has no channel to close. On wake a waiter re-runs the warm path rather than trusting the signal — the model may have been evicted in between. A waiter whose client disconnects returns immediately and the job keeps running; it belongs to the job record, not to the request. A failure is recorded on the row so every waiter reports the real cause, and the row survives briefly so the next request does not read "no job" as "not loading" and start a duplicate load of a model that just failed. The runner heartbeats the row on a fixed interval whether or not bytes are moving, which is what keeps a legitimately silent checkpoint load from being reclaimed as an orphan. Phase (installing/staging/loading) and placement ride to the heartbeat on the context, the same seam load_deadline.go already uses, so single-host paths are untouched. Non-distributed mode (no DB) keeps the inline load exactly as it was. Assisted-by: Claude Opus 5 [claude-code] * feat(distributed): bound the wait for a loading model and answer with progress A request whose model is cold-loading now attaches to the running job and is served the moment the model is ready. That wait has to be bounded: a held HTTP request cannot survive real infrastructure, and an ingress or LB idle timeout kills a twenty-minute request regardless of what LocalAI does. New LOCALAI_MODEL_LOAD_WAIT (default 60s) bounds the CALLER, never the load — the job keeps running either way. On expiry the request gets 503 with Retry-After and a structured body naming the model, the node, the phase, byte progress and an ETA. The `error` envelope keeps OpenAI clients working; `loading` is additive so they ignore it. The ETA comes from the job's own observed rate and is omitted rather than guessed until enough bytes have moved for that rate to mean anything: a confidently wrong ETA on a twenty-minute wait is worse than none. Retry-After is that ETA when known, clamped to [5s, 300s], and the wait budget otherwise. LOCALAI_MODEL_LOAD_WAIT=0 waits unbounded, for deployments with no proxy in front. Zero in the config struct still means "unset, use the default", so the CLI records the operator's zero as ModelLoadWaitUnbounded rather than losing the distinction. The distributed branch of ModelLoader.loadModel wrapped the router's error with %s, which flattened it to a string. Use %w: the typed error is what the HTTP layer keys the 503 off. Assisted-by: Claude Opus 5 [claude-code] * feat(api): add GET /api/models/{id}/load-status A client that receives 503 while a model stages onto a worker needs somewhere to poll. This returns the same `loading` object the 503 carries — phase, node, byte progress and ETA — or 404 when no load is running. Read-only and observability-shaped, so it is deliberately neither admin-gated nor feature-gated: it explains a 503 the caller just received, and hiding that behind a per-modality feature would make the explanation for a failed image request depend on chat permissions. It also gets no MCP tool, since there is nothing here an admin would manage conversationally. Registered on the surfaces from .agents/api-endpoints-and-auth.md: the swagger block (existing `models` tag, so /api/instructions needs no new area), the endpoint discovery maps in RegisterLocalAIRoutes, regenerated swagger, and the distributed-mode docs page. No FLAG_* usecase is involved, so capabilities.js is unchanged. Assisted-by: Claude Opus 5 [claude-code] * feat(ui): show cold-load progress in Chat and retry when the model is ready A chat request for a model that is still staging onto a worker now gets a 503 carrying live progress instead of an error. Render it: the composer shows the phase (installing / staging / loading), the node, the percent and the ETA, then polls load-status and re-sends the request the moment the model is ready. Reuses the staging progress idiom the page already had rather than inventing a second one — the two sources are folded into one loadProgress, with the load job winning because it is authoritative across frontend replicas and knows the phase, where the staging operation only knows about a byte transfer this replica happens to be performing. Waiting is bounded (three send attempts, ~30 min of polling each), so a load that never finishes still surfaces as an error rather than as a spinner nobody questions. An aborted generation stops the polling too. Assisted-by: Claude Opus 5 [claude-code] * fix(distributed): check warm-path cleanup errors The router moved legacy cleanup calls onto newly linted lines. Report cleanup failures while preserving the fallback to a cold load. Assisted-by: Codex:gpt-5 [golangci-lint] --------- Co-authored-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com>	2026-08-15 13:20:11 +02:00
localai-org-maint-botandlocalai-org-maint-bot	133c546c3f	feat(api): add text moderation endpoint (#11316) ... * feat(api): add text moderation endpoint Add an OpenAI-compatible /v1/moderations endpoint backed by constrained local text generation. Register its auth and discovery surfaces, document the text-only MVP, and cover response shaping and access control. Assisted-by: Codex:gpt-5 * test(mcp): update assistant client stub Keep the LocalAI Assistant holder test stub aligned with the scheduling methods added to LocalAIClient so repository-wide type checking succeeds.\n\nAssisted-by: Codex:gpt-5 --------- Co-authored-by: localai-org-maint-bot <306269227+localai-org-maint-bot@users.noreply.github.com>	2026-08-03 18:03:46 +02:00
mudler's LocalAI [bot]andEttore Di Giacinto	5cb0c1a872	feat(ui): close the gap between the shipped UI and the design mocks (#11307) ... * feat(ui): give the Operate overview real numbers and traces a latency shape First two items from a component-by-component comparison against the mocks. The pattern that audit found: everything newly built matched, everything pre-existing got the palette but not the layout, and an "absent rather than empty" rule hid most of the overview exactly when someone was looking at an idle installation. **The headline grid is always rendered**, including at zero, with a fourth cell for host memory. Hiding it removed the page's structure precisely when it was most likely to be read, and "0 failed" is information — an absent panel is not. The quiet case is now said in a line underneath instead of by showing nothing. **The sections state counts** rather than listing their destinations: backends, models, updates and running operations instead of the words "Usage and traces". That needed installed backend and model counts in the summary context, which are two more cheap reads on the poll that was already running. **Traces rows carry latency as a bar as well as a figure**, scaled against the slowest request currently in view and turning amber past two seconds. The table had no latency column at all — the number was buried in the expanded detail, so the shape of the tail was invisible while scanning. Scaling against the view rather than an absolute ceiling is deliberate: what matters when reading a page of traces is which of these are the outliers, and an absolute scale flattens every row on a fast installation into nothing. Full e2e suite: 409 passed, 4 skipped. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5[1m] [Read] [Edit] [Bash] * feat(ui): name the engine on Home's resident models, and add jump-back-in Third item from the mock comparison. The mock showed each resident model with the engine serving it. /system carried only the id, so the audit recorded this as blocked on a server field — but the config loader is already in scope where that response is built, so it is one lookup. SysInfoModel gains an optional `backend`, resolved from the model's config and omitted rather than guessed when there is none (a loose file, or a config since removed). Home renders the column blank in that case; the test pins both halves of that. Memory per model stays out. It is not one lookup — it would mean asking each backend process — and inventing a number beside a real one is worse than leaving the column off. "Jump back in" is the block the mock had and Home did not. The quick-links row above it is a set of first-run actions; these are the three places someone returns to, each stated with what it currently holds rather than as a bare label. Go: routes suite passes. Full e2e suite: 412 passed, 4 skipped. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5[1m] [Read] [Edit] [Bash] * feat(ui): rank the recommended models as lanes instead of equal cards The hardware recommendations were a grid of equally-weighted cards. The list is already sorted by fit, and a grid throws that order away: three cards side by side say "pick one", when the page has actually formed an opinion about which one. They are lanes now, read top to bottom in fit order, with the leader carrying the single amber "Best fit" label and the rest marked "Also fits". One opinion per page — the alternatives are alternatives, not runners-up each worth their own colour, which is how a strip of coloured badges ends up meaning nothing. Below 720px the size and VRAM columns drop and the lane keeps the name and the install action, which are the two things a narrow screen needs. The existing panel spec moves off .rec-models-item onto .lane rather than being deleted; dismissal, collapse, keyboard operation and install all still pass unchanged, and there is a new assertion that exactly one row is called out. Full e2e suite: 413 passed, 4 skipped. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5[1m] [Read] [Edit] [Bash] * fix(ui): drop capsule chips app-wide, and un-break the empty voice library **Pills are gone.** A capsule radius reads as a tag floating on the surface, which fights a system whose structure is hairlines and square corners — and with chips on Discover, Host, Activity and the biometrics pages, "some pages have pills" was the real inconsistency rather than any one page. Sixteen selectors move to the small radius: filter buttons, tab pills, activity and biometrics chips, file and count badges, the jump-to-latest control, the nav badge. Round *buttons* keep their circle — .lightbox__nav and .home-send-btn are circles, not capsules — as do every progress track, status dot and avatar, which are round because they are round, not because they are tags. **The empty voice library was unusable.** `.voice-library-empty` sets min-height: 430px, border: 0 and background: transparent — a description of the empty PANEL — and it had been attached to the action instead. The create button was therefore a 430px transparent box that pushed itself out of the panel and could not be seen. Moved onto the container it describes, which now centres its action rather than letting it fall off the bottom. Same class-mangling shape as the Agents header fixed earlier. Full e2e suite: 416 passed, 4 skipped. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5[1m] [Read] [Edit] [Bash] * feat(ui): put Host's headline figures on the shared hairline strip Host had shadowed, clickable StatCards above a page that already has a rail, a pane and a tab bar — a second dashboard language on one screen, and a different one again from the figures inside its own detail pane. The Operate overview's figure grid is generalised into a shared `.stat-strip` and Host adopts it, so the two pages read as one system: same cell, same figure scale, same tone vocabulary, and the same hairline grid the split-view StatGrid already uses. The cells stay clickable and still route into the tab and filter they describe, because a count is worth more when it is also the way to the thing counted. Tone is spent only where the number means something — running and updates when non-zero — since a strip where every cell is coloured has no emphasis left. Two bugs made on the way, both now covered: - The first version put `<button>` elements inside a `<dl>` with `<dt>`/`<dd>` inside the buttons. Neither is valid, the browser re-parents both, and the cells collapsed. These cells are a set of controls, so a plain container of buttons is also the honest markup. - Even correct, the strip rendered 2px tall: `.page--app` is a flex column whose split view takes flex:1, so a child with no intrinsic minimum is shrunk away. The old cards survived only because `.stat-card` carried min-height:96px. The strip now declines to shrink, with a test pinning it. The stat-card specs are retargeted rather than deleted: they were written to guard a class collision on a page that no longer uses cards, so they now guard the strip's labels and its height. Full e2e suite: 417 passed, 4 skipped. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5[1m] [Read] [Edit] [Bash] * feat(ui): make Backends notices an edge rather than a filled card The install and upgrade banners were tinted cards with a full border. A filled panel makes every notice shout at the weight of an error, which is how notices stop being read — and Backends shows one on most visits, so it was shouting routinely. They are now a hairline with a coloured left edge, the same treatment the Operate overview gives rows that want a decision, so "this needs you" looks the same wherever it appears. Counts in the notice take the monospace tabular figures the rest of the console uses. Also drops the last inline style on the page, and refreshes the inline-style baseline, which has read 624 against a real count since #11288 landed. The gate exits 0 either way, so nothing was failing — but a baseline 86 above the truth would have let that many inline styles back in unnoticed. Now at 538, which tightens the ratchet rather than loosening it. The spec creates the upgrade it asserts on rather than skipping when the mock has no notice: a test that skips is a test that proves nothing. Full e2e suite: 418 passed, 4 skipped. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5[1m] [Read] [Edit] [Bash] * feat(ui): finish the mock parity list, and stop hiding the recommendations The last two items from the audit, plus a correction. **Discover's use-case shelf is lanes.** These are a list of ways in, read in order; a grid of equal cards asks the reader to compare them, which is not the choice on offer. **The request panel reaches every generator.** Video, 3D, Sound and Audio FX join Images and Speech, so each one teaches its own endpoint rather than two of six doing it. Audio FX records the fields that shape the request rather than the bytes, since its payload is multipart. **Recommendations no longer collapse themselves.** They were folded away by default once anything was installed. That is the page's one opinion about this host, and an opinion hidden by default is one the reader never gets. Someone who disagrees can still collapse it and that choice is remembered — the difference is that we no longer make it for them. Three specs asserted the old default and now assert the new one. The use-case heading also sat a line's width from the text it introduces, so the two read as one paragraph. It has air under it now, and the shelf is separated from the recommendations above it. Two tests removed rather than kept: a generator loop whose only real assertion was `expect(endpoint.length).toBeGreaterThan(0)`, and an earlier card-gap guard that could only skip. A test that cannot fail is worse than no test, because it reads as coverage. Full e2e suite: 418 passed, 4 skipped. Inline styles at baseline. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5[1m] [Read] [Edit] [Bash] * fix(ui): make the Host figures legible and give the strip its spacing back Three defects introduced by the Host redesign, all found by looking at the running app rather than by the suite. **The figures were invisible.** "Running now" and "Updates available" rendered pure black on the dark ground. Two causes compounding: the `--muted` tone alias never landed, because the source rule has extra spaces before its brace and the exact-match edit missed it silently; and a `<button>` does not inherit colour, so with no tone rule the value fell back to the user agent's `buttontext`. Both fixed, and a test now fails on any figure computing to pure black. **The strip sat flush against the resources panel.** `.stat-strip` declares `margin: 0 0 ...` and is declared later in the file than `.manage-summary`, so the shorthand quietly won and the top margin became zero. Raised to `.stat-strip.manage-summary` so it beats the shorthand on specificity rather than on declaration order, which is the kind of thing that breaks again the next time a rule moves. **Discover's use-case heading had a doubled gap.** `.zero-pane` is a flex column that already separates its children; adding a margin on top of the gap stacked the two. The margin is gone and the heading keeps only its own breathing room. Full e2e suite: 420 passed, 4 skipped. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5[1m] [Read] [Edit] [Bash] * feat(ui): make Studio's tabs path segments rather than a query parameter `/app/studio?tab=images` reads like a filter applied to a page. It is navigation: a different generator, with its own state and its own deep link. It is now `/app/studio/images`, with the overview at `/app/studio`. Legacy `?tab=` links are redirected once to the path form, replacing the history entry so Back does not bounce between two spellings of the same place. Bookmarks and older links keep working and land on the canonical URL rather than a second version of it, which is the part worth having a test for. The nine `?tab=` references were all in specs, none in docs, so the migration is contained. They move to paths, and a new spec pins the redirect. Full e2e suite: 421 passed, 4 skipped. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5[1m] [Read] [Edit] [Bash] * feat(ui): make the hardware recommendation a section, not a dismissable card It was a bordered card with a collapse control and a close button, sitting inside a pane that is otherwise hairline sections. Two problems: it read as something bolted onto the page rather than part of it, and treating it as an interruption to be shut is the wrong frame for the one thing the page has to say about the machine it is running on. It is now a plain section with the same heading treatment as the shelves below it. The collapse state, the dismissal, their storage keys and the legacy key read for backwards compatibility all go with it, along with the installedCount prop that existed only to pick a default collapse. Five specs described behaviour that no longer exists and are removed rather than adjusted — collapsing, dismissing, persistence of both, and the toggle's keyboard handling. One new spec asserts the replacement contract: no control with aria-expanded, no dismiss, and no card border. Full e2e suite: 416 passed, 4 skipped. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5[1m] [Read] [Edit] [Bash] * fix(ui): restore every stripped icon and every default-chrome control You reported two broken icons. They were not two: an earlier automated edit had stripped the `fa-*` class from twenty `<i>` elements across eleven pages, and an `<i>` with no icon class renders nothing at all. Settings' save button, the voice-profile back link, and eighteen others — agent row actions, task and job buttons, import and create actions — were all drawing empty space. Each is restored from its own context rather than a blanket icon: the agent row gets pause/play, pen, comments, file-export and trash; the fine-tune toggle swaps plus for xmark as it opens; the P2P documentation link gets the external-link glyph. The same edit left controls without their classes. Fine-tune's "Import config" was rendering in the browser's own chrome, and `.p2p-cmd__copy` set a border but no background, so it fell back to `buttonface` — a pale grey chip on a dark command block. FineTune's "New job" also had its icon classes folded into the button's className, the same mangling already fixed on the Agents header. Rather than fix the reported two and wait for the next report, this adds a standing audit: twenty-five routes are walked and the test fails on any visible control rendering with user-agent chrome, or any `<i>` without an `fa-*` class. It found the three remaining cases after the first sweep, and it is the reason the next one cannot ship quietly. Full e2e suite: 417 passed, 4 skipped. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude Code:claude-opus-5[1m] [Read] [Edit] [Bash] --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-08-03 12:16:09 +02:00
Adiraandlocalai-org-maint-bot	ef724a3c9d	feat(api): add /v1/detokenize endpoint (#9620) ... * feat(api): add /v1/detokenize endpoint Closes #1649. Mirror of the existing /v1/tokenize path, requested by @benniekiss in the issue thread for "complete API workflow" use cases that need to turn token IDs back into text without local processing. - Add Detokenize gRPC RPC with DetokenizeRequest{tokens} / DetokenizeResponse{content} messages. - Implement in the llama.cpp backend using common_token_to_piece, the same primitive TokenizeString already uses internally. - Other backends inherit the default Unimplemented from base.Base, in line with how Detect, Rerank, etc. are gated per-backend. - Wire up the Go gRPC interface, server, client, and in-process embed wrapper alongside their TokenizeString counterparts. - Add the schema types, ModelDetokenize wrapper, HTTP handler, route registration, RouteFeatureRegistry entry (gated by FeatureTokenize so no new feature flag is needed), and the discovery map entry under ai_functions. - Regenerated swagger reflects the new endpoint and types. - Update authentication.md to list /v1/detokenize alongside /v1/tokenize. Assisted-by: Claude:claude-opus-4-7 Signed-off-by: Adira Denis Muhando <dennisadira@gmail.com> * test(e2e): add mock backend tests for /v1/detokenize Add Detokenize to the mock gRPC backend and wire up two e2e tests in the MockBackend suite: one that posts known token IDs and asserts a non-empty content response, and a round-trip that tokenizes first then detokenizes the returned IDs. Addresses reviewer feedback on #9620. Assisted-by: Claude:claude-sonnet-4-6 Signed-off-by: Adira Denis Muhando <dennisadira@gmail.com> * fix(kokoros): implement detokenize in the Rust backend service The Detokenize RPC added in this PR grows the tonic-generated Backend trait. Unlike the other languages there is nothing to inherit a default from — Rust trait impls must list every method — so backend/rust/kokoros failed to compile: error[E0046]: not all trait items implemented, missing: `detokenize` --> src/service.rs:72:1 72 | impl Backend for KokorosService { Go backends pick up the Unimplemented default from base.Base, and the generated C++/Python servicer bases default to UNIMPLEMENTED, which is why the Rust backend was the only one that broke. kokoros is the sole Rust crate in the tree, so this is the full extent of the fallout. Return Status::unimplemented("Not supported"), matching how this same file already gates tokenize_string and ~20 other unsupported RPCs. Fixes the tests-kokoros and backend-jobs-singlearch-4 (-cpu-kokoros) failures on the previous head. Assisted-by: Claude:claude-opus-5 cargo Signed-off-by: Adira Denis Muhando <dennisadira@gmail.com> --------- Signed-off-by: Adira Denis Muhando <dennisadira@gmail.com> Co-authored-by: localai-org-maint-bot <bot-opensource@localaisrl.com>	2026-07-30 16:01:47 +02:00
mudler's LocalAI [bot]andEttore Di Giacinto	9c85cacfe3	feat(audio-cpp): add the audio.cpp native backend (#11141) ... * backend(audio-cpp): add the native build scaffold Links 0xShug0/audio.cpp engine_runtime through its public framework headers and serves Health/Status. Model loading and the audio RPCs follow. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): keep the build-tree rpath at $ORIGIN Upstream sets CMAKE_BUILD_WITH_INSTALL_RPATH in its own directory scope, so CMake was appending its build-tree library dir to our target and baking an absolute build-host path into the shipped binary. Set BUILD_WITH_INSTALL_RPATH on the target so a package that forgets to bundle libggml*.so fails on the build machine too, instead of only on a user's box. Also document why EXCLUDE_FROM_ALL must stay on the add_subdirectory call, correct the claim that Ubuntu ships no gRPC CMake config, stop the pin comment from repeating the assignment token that bump_deps.sh rewrites, and make test-engine fail rather than pass when no test is registered. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): parse namespaced model options Splits option entries on the first colon so path values survive, and routes load./session. prefixes to the upstream load and session option maps. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): reject out-of-range numeric model options std::atoi is undefined once the digits exceed long and in practice wraps, so device:2147483648 was accepted and handed the ggml backend selector a device index of -2147483648 from a function whose error text promises a non-negative integer. Parse with strtol and reject on ERANGE, on a value above INT_MAX, and on any unconsumed trailing input. The error strings are unchanged. Name the whole entry in the unknown-key error too: an entry like ':value' has an empty key and left the user nothing to grep for in their YAML. Tests look keys up through a helper instead of map::at, so a prefix off-by-one fails one named check rather than aborting the binary and skipping the rest of the suite, and cover the overflow, negative and non-numeric paths. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): route LocalAI RPCs onto audio.cpp tasks Task-major resolution over the family's advertised capability set, with the voice-reference and instructions signals selecting cloning and voice design, and a streaming-to-offline fallback for server-streaming transcription only. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): use upstream's 'spk' task name and pin the preference order The SpeakerRecognition short name was 'spkrec', which audio.cpp neither prints nor parses; a name copied out of audio.cpp was rejected and a pinned 'spkrec' would not survive the engine boundary. Emit 'spk', keep 'spkrec' as an input-only alias, and correct the known-tasks lists. Three assertions were vacuous because their fixtures advertised a single task, so reversing a preference order or dropping the RPC name and the attempted pairs from the capability error all passed. Give them fixtures that can tell the orderings apart. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): convert sample, time and PCM units Integer nanosecond conversion so 44.1 kHz stays exact, float seconds for the VAD and diarization messages, and saturating s16le encode so an overshooting sample cannot wrap to the opposite sign. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): harden seconds_to_samples against NaN and overflow seconds_to_samples is the one entry point fed by untrusted-shaped input: a float-seconds timestamp off the wire, or a boundary from a model that diverged. Its guard covered only the low side, so NaN and out-of-range values fell through to an undefined double-to-int64 cast and came back as INT64_MIN. A hugely negative sample index used later as an offset or a length is a wild pointer rather than merely a wrong timestamp. Reject NaN with the !(x > 0) form and saturate before the cast. Also round instead of truncating there. These functions exist to cross the float seconds boundary the VAD and diarize messages use, and truncation lost a sample about half the time on the samples-to-seconds-and-back round trip, starting at n=1. Pin the decode scale at INT16_MIN, pin nanosecond truncation on a nonzero fraction, and record why the clamp argument order in f32_to_s16le is load-bearing for NaN. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): map NaN PCM samples to silence explicitly f32_to_s16le relied on std::min argument order to keep a NaN sample away from std::lround, whose result is unspecified for NaN. That was too subtle to rest on a comment, and the comment was itself wrong: it warned against a spelling that the outer std::max already catches, while three real spellings leak, including std::clamp, which is the idiomatic C++17 way to write the same clamp and so the likeliest future edit. Divert NaN before the clamp and encode it as 0. A NaN sample rendered as a full-scale click is worse audio than a dropped one, and this unit converts audio that may have originated off the wire. Pin it with an exact-value check rather than a range check, since all three outcomes the plausible spellings produce are finite and inside full scale, plus an invalid-operation check that fails unless the NaN is diverted before any ordered comparison. That second check is what catches modernizing the clamp and dropping the guard together. Also bound the seconds round-trip comment, which claimed unconditionally what holds only below roughly 2^23 samples, and document NaN, saturation and that bound in the header. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): assemble transcripts from runtime spans The top-level transcript text is TaskResult.text_output verbatim. audio.cpp carries text nowhere else: speech_segments, speaker_turns and word_timestamps hold spans and labels only, so deriving the text from them empties the transcript for any producer that omits word timing, VibeVoice diarized ASR included. Fixtures cover every observed producer shape. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): keep a nested speaker turn's own label A segment sourced from speaker_turns re-derived its speaker by greatest overlap. A turn's overlap with its own span is the largest possible, so a turn nested inside another speaker's turn could only tie with the container, and the tie went to whichever came first. sortformer_diar binarizes each speaker independently and sorts by start sample, so the container always comes first and the interjecting speaker was silently erased from DiarizeSegment.speaker. choose_segment_spans now carries the label out with the span. Also pins the nearest-segment fallback against measuring from either endpoint or from segment position, which a trailing-only stray word could not do, and exercises the empty-word guard in join_words. Two fixtures that pin a rule but do not mirror any pinned family are relabelled defensive. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): serialize runs with a wedge-aware guard audio.cpp sessions are not reentrant and a wedged CUDA call cannot be cancelled, so a plain mutex would pile every worker thread behind a stuck GPU. Callers waiting past the configured bound, or arriving while the holder has already overrun it, fail fast instead. A caller that queues behind a healthy run deliberately does not stamp the clock: only the thread that takes the lock does. Stamping on arrival would restart the wedge clock on every request and hide a stuck run from everyone behind it, which is the pile-up this guard exists to prevent. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): serialize inference through an InferenceLane One audio.cpp model is loaded per backend process and its sessions are not reentrant, so concurrent gRPC handlers have to take turns. Serialization alone is not enough: a wedged GPU call cannot be cancelled from userspace, so an unbounded queue behind one stuck run would swallow every gRPC worker thread until the process is useless. InferenceLane gives handlers a lane with room for one runner. LaneEntry occupies it for a scope and gives it back on every exit, including an exception, and is the only way to take the lane at all: occupy/vacate are private with LaneEntry as the sole friend, so a caller cannot acquire without holding something that releases. LaneEntry is immovable on purpose, because a moved-from entry would have to stop releasing while the lane still recorded it as occupied. A caller either waits indefinitely or brings a millisecond budget. A bounded caller that cannot get in fails instead of waiting on, and a bounded caller whose budget is already shorter than the age of the run in the lane fails immediately, which is what stops a queue forming behind a wedged run. The two failures carry different text: one names the wait it exhausted, the other states the measured age of the run without claiming to know why it is long, since a short budget meeting a legitimately long run lands there too. The run's age is stamped only after acquisition. A waiter that published itself as holder would restart the measurement and hide a genuinely stuck holder from every caller behind it. Budget negotiation and the overrun decision are pure functions taking their inputs explicitly, so both are covered without threads or sleeping. The per-model ceiling arrives as an int of milliseconds; a request may tighten it and may never loosen it. Replaces the previous run_guard unit, which was a derivative of an Apache-2.0 file upstream and could not stay in an MIT tree. Written from a behaviour contract with no reference to the removed code. Tests: 65 checks, standard library only, single translation unit, clean under -Wall -Wextra. Mutation tested at 23/23 killed; two of those mutants exposed missing coverage and the tests were extended until they died. ThreadSanitizer clean. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): make the B10 test able to fail, and document LaneEntry Review of the previous commit found the B10 test could not fail for the reason it was named. It aged the in-flight run to about 120 ms and then tried two budgets, 30 ms and 60 ms, both under that age, so both callers took the fail-fast path. "The two failure modes do not share one message" was comparing two fail-fast messages that differ only in the budget they print, and the timeout path was never reached. The second budget is now 400 ms, well over the run's age, so that caller queues and times out, and a new check asserts which path each caller took instead of inferring it from inequality. A mutant that makes the fail-fast path emit the timeout message previously died only on B4 and B8 checks; it now also dies on B10. Comment-only changes elsewhere. LaneEntry now says it is not reentrant and does not detect reentrancy: a second entry on a thread that already holds the lane surfaces as LaneUnavailable with a positive budget, but parks silently in unbounded mode, which matters because a handler may hold one across a whole stream. The immovability note now names the shapes that work, an optional emplaced in place or a unique_ptr, rather than saying to hold the entry indirectly without saying how; all three documented forms were compiled before being written down, which is how the note came to say that an optional of an immovable type cannot itself be returned. The header's explanation of why fail-fast exists is reworded. Two clauses traced back to a specification written after reading the Apache-2.0 upstream header, and while that was judged de minimis, this unit was rewritten precisely to carry no upstream expression at all. The margin table in the report was also wrong about which wall-clock margins are load-sensitive: there are four, not one, and the tightest is the B3 arrival check, which is now flagged at the call site. No margin value changed and none moved across 65 runs. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): gate model loading on the audio.cpp family Refuses any GGUF without an audiocpp.model_spec.family key and any non-GGUF path without an explicit family option, so the model loader's greedy backend probe cannot bind an unrelated llama.cpp GGUF to this backend (#9287). Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): load models and cache sessions per task Loads one ILoadedVoiceModel and creates an IVoiceTaskSession lazily per (task, mode), so the same model serves both the unary and streaming RPCs. LoadModel derives the family from GGUF metadata or an explicit option and fails with INVALID_ARGUMENT otherwise, so a failed load is a gRPC error the backend probe can see. audiocpp_backend::Task mirrors engine::runtime::VoiceTaskKind positionally, and drift there is silent: every unit still compiles and every test still passes while the backend runs a different task. Two mechanisms pin it. The static_asserts in loaded_model.cpp catch an insertion or a reorder, and -Werror=switch on that one file turns an appended upstream enumerator into a build failure rather than a warning in a 600 file log. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): stop aborting the process on SIGTERM The signal handler called grpc::Server::Shutdown directly. Shutdown takes an absl::Mutex, which is not async-signal-safe: the handler can interrupt a thread already holding that mutex, and abseil's deadlock detector responds by aborting. Every SIGTERM therefore ended in exit 134 and a 'dying due to potential deadlock' stack rather than a drained shutdown. The handler now sets a lock-free atomic and returns. Server::Wait moves to a helper thread so the main thread can poll that flag and call Shutdown itself, outside any signal context. A condition variable would not have helped, because notifying one from a handler is not async-signal-safe either. SIGTERM and SIGINT both exit 0 with no stack trace, where both previously exited 134. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): correct the status, lifetime and state contracts of LoadedModel An environment fault during session creation was reported as UNIMPLEMENTED. A missing libggml-cpu-*.so surfaced to the client as 'family silero_vad advertises vad/offline but refused to create the session: Failed to initialize CPU backend', which tells LocalAI the model cannot do this and must never be retried, and sends an operator hunting a capability bug instead of a packaging one. A throw from create_task_session is now a plain runtime_error, so it maps to INTERNAL. Only a null return, where the family genuinely declined, stays a CapabilityError. The model.'s task: option was parsed and then dropped: it lived in a local that died at the end of LoadModel and had no route to RequestShape::pinned_task. LoadedModel now keeps it and exposes pinned_task(). The global model becomes a shared_ptr reached through snapshot(). An audio RPC runs for seconds and cannot hold g_model_mu for its duration, so under a unique_ptr a Free arriving mid-request would destroy the model underneath it. Handlers now take a counted reference and whichever finishes last does the teardown, outside the lock. session_for documents the streaming state contract rather than resetting the session itself. Resetting on a cache hit was tried first and is not possible: silero_vad throws 'session prepare() must be called before Silero VAD reset()', so it would turn an ordinary second fetch into a hard error. start_stream's base implementation is already a reset, so a caller that runs prepare then start_stream per stream gets a clean session; a probe against the bundled silero_vad confirms an identical replay when it does and a carried-over stream when it does not. Also: an unknown backend: name is rejected before the model loads rather than after; MainGPU is parsed instead of passed through std::atoi, which turned 'gpu1' into device 0 silently; and device carries a device_set flag, because 0 is both the default and a real device index, so MainGPU was overriding an explicit device:0 that the neighbouring threads: handling promises will win. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): serve the VAD and Diarize RPCs Both emit float seconds, converted from the runtime's sample-index spans, and both take a counted reference to the loaded model through snapshot() and hold it for the whole call: a Free arriving mid-request drops only the global's reference, so whichever request finishes last destroys the model instead of one of them running on freed weights. An AddressSanitizer build reproduces exactly that heap-use-after-free inside ggml_vec_dot_f32 when the handler keeps a raw pointer instead, which is why the shape is what it is. The inference lane is taken before session_for, not after. session_for reads and writes an unsynchronised session cache and the offline run calls prepare(), which mutates the session, so both belong inside the lane. Diarize routes before it reads the input file, so a family that cannot diarize at all says so rather than complaining about the audio first. Its per-segment text stays empty because audio.cpp's SpeakerTurn carries a span and a speaker label only, and nested or overlapping turns are passed through untouched: a sortformer turn inside another speaker's turn is correct output for overlapped speech, and LocalAI is overlap-tolerant downstream. Duration counts frames rather than floats, so a stereo input does not report twice its length. Verified end to end against upstream's bundled silero_vad, which needs no download, using the bundled 16 kHz speech asset: a synthetic tone returns nothing, correctly, because silero detects speech and a sine is not speech. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): enforce ModelIdentity on VAD and Diarize audio-cpp was the only C++ backend without the model-identity guard, and no later task in the plan added it. pkg/grpc/server.go enforces checkModelIdentity on exactly these two RPCs, for the reason #10952 records: in distributed mode a worker can recycle a stopped backend's gRPC port for another model's backend, and the controller's liveness-only probe cannot tell a stale cached route from a live one. Without this guard a stale route gets a different model's VAD or diarization answer back with a 200. The loaded identity lives on LoadedModel rather than in a separate global, which is where this differs from llama-cpp. A handler holding the model through snapshot() then necessarily judges against the identity that model was loaded with, and a concurrent reload cannot swap one without the other. The refusal is NOT_FOUND carrying the verbatim grpcerrors.ModelMismatchSentinel substring. session_for and run_offline now take a const LaneEntry & proof-of-holding parameter. The rule that both must run under the inference lane was prose, which is exactly how the plan came to specify the inverted order; it is now a compile error. Restoring the inverted order fails to build rather than racing on an unsynchronised session map with a mutating prepare(). Diarize's speaker-hint comment claimed the dropped hints were "not a silent failure". From the caller's side that is what they are, and backend.proto documents num_speakers as forcing, so the comment now says plainly that the forwarding is dead for sortformer and that the family which lands must either honour num_speakers or refuse it. read_audio_file inspects the error_code from exists(), so an unsearchable parent directory no longer reports as a missing file. The VAD handler records the stimulus that actually works, since silero correctly ignores synthetic tones and the next task would otherwise rediscover that. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): make the lane and identity guards structural Two hardenings ahead of the eleven handlers still to be written, both of which get harder to retrofit later. The lane proof-of-holding parameter was a const reference, which binds to a temporary, so session_for(rpc, shape, model->acquire(0)) compiled. Each such temporary dies at the end of its own full-expression, releasing the lane between two calls that must share one: precisely the split the parameter exists to prevent, and the form a future author is most likely to reach for because it reads as tidy. A non-const reference requires an lvalue, so the temporary form now fails to compile while the named-local handlers build unchanged. The header comment no longer implies the check is total either: it proves a lane was taken, not that it is this model's lane. The identity check was two lines each handler had to remember, with nothing failing if a new one forgot them and no C++ equivalent of model_identity_modalities_test.go to notice. snapshot() becomes snapshot_unchecked(), whose only legitimate caller is Status, since HealthMessage carries no ModelIdentity. Handlers go through snapshot_for(), which takes the counted reference, refuses when nothing is loaded, and runs the identity check before anything can route. Every handler already has to call something to obtain the model, so the guarded call is now the shortest path and skipping it means deliberately typing snapshot_unchecked. A convention that has to be remembered can rot; this cannot. Verified: the temporary-argument and inverted-order forms each fail to compile with the expected diagnostic, the real handlers build, and bypassing the guard in Diarize alone turns the identity test red on that RPC while VAD stays green. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): serve the AudioTranscription RPC Adds result_map, the engine-to-proto boundary, and wires the offline transcription RPC. The handler branches on the ROUTED task: for Asr the request's prompt is whisper-style decoding context and becomes a request option, for Alignment the same field IS the transcript to align and becomes the text input. Routing has already decided which. The result text is TaskResult.text_output verbatim and is never derived from the segments. audio.cpp carries transcript text in text_output and nowhere else, so deriving it returns an empty transcript for every producer that reports segments without word timing. transcript_assembly already enforces that; this commit's job is not to undo it at the proto boundary, and result_map_ctest pins it there. read_audio_file now takes the sample rate the caller needs. Both file-fed speech handlers ask for 16 kHz mono, for two reasons: silero_vad and sortformer_diar refuse anything else outright, which turned an ordinary 44.1 kHz upload into INTERNAL, and nemotron_asr emits word timestamps in its own 16 kHz feature domain whatever the input was, so only a 16 kHz buffer makes the emitted nanoseconds right. Zero keeps the file's native rate and channels, which is what source separation will need. LoadedModel::check_can_serve answers a capability refusal before the lane is taken and before the input file is read. Routing is a pure read of the immutable capabilities, so a model that cannot serve an RPC no longer waits out somebody else's run to say so. VAD and Diarize use it too. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): stop linking sentencepiece's vendored protobuf engine_runtime links sentencepiece, whose default SPM_PROTOBUF_PROVIDER builds the protobuf-lite 3.14.0 sources it vendors. The generated backend.pb.cc is built against the toolchain's protobuf 3.21.12. Both ended up in the binary: 476 google::protobuf:: symbols came from the archive, 278 of them also defined by libprotobuf.so, and the archive won, because once ld pulls a member in for sentencepiece's own code every reference binds to the definitions that member carries. The visible symptom is one function. ParseContext::ParseMessage(MessageLite*, const char*) is what a generated _InternalParse calls for a submessage field and for nothing else, so flat messages parsed and nested ones did not: a TranscriptResult carrying segments serialized to correct bytes that the same process could not read back, and TranscriptLiveRequest, a oneof of submessages, could not have been parsed at all. Underneath that, 3.21 generated code was running 3.14 arena, ArenaStringPtr and ExtensionSet code. -Wl,--exclude-libs does not fix it. It makes those symbols LOCAL in .dynsym and the parse still fails, because the binding was decided at static link time and no visibility flag revisits it. Setting SPM_PROTOBUF_PROVIDER to "package" before add_subdirectory points sentencepiece at the protobuf the generated code was already built against. Zero google::protobuf:: definitions remain in the executable afterwards, every nested message round trips, and citrinet_asr, which parses a SentencePiece ModelProto at load time and would break first if this were wrong, still tokenizes and transcribes correctly. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): fix the segment text a transcription response is built from Segment text is not decoration. core/http/endpoints/openai/transcription.go routes response_format text, srt, vtt and lrc through schema.TranscriptionResponse, which builds the entire body out of Segments[].Text and never reads the top-level text. So for those four formats the segment text IS the response. nemotron_asr emits one word_timestamp per SentencePiece token, and the word boundary is carried as a LEADING SPACE on the piece ("So", "me", " call"). join_words inserted a space unconditionally, so response_format=text returned "So me call me na ture ," while the correct sentence sat unread in the top-level field. The separator is now chosen from the words themselves: whole words are space-joined, subword pieces are concatenated, and one leading space anywhere selects the latter. Concatenating the real nemotron pieces reproduces text_output exactly, verified end to end. This does not touch the top-level text, which is still text_output verbatim. The rule that forbids deriving the transcript from the segments is about the direction segments -> text; segment text has no source other than its words. Two smaller corrections in the same area: timestamp_granularities ["word"] set only "word_timestamps", a key no family in the pinned upstream reads. It now sets "return_timestamps", which qwen3_asr does read and which both runs its forced aligner and shortens its chunk window, so asking for word granularity no longer silently returns nothing. The request-option comment claimed more than it delivered. prompt, translate and temperature are read by no ASR family, and are forwarded only so a family adopting them works unchanged; the comment now says so per key, and gives TranscriptRequest.diarize the same explicit treatment threads already had. Also: the shipping target now carries -Wall -Wextra -Wpedantic, which it never did, so "the build is clean" starts meaning something; and fill_transcript_result no longer swallows a null response pointer, since answering OK with an empty transcript is the one failure mode this unit exists to prevent. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): serve the AudioTransform RPC Covers voice conversion, singing voice conversion, speech to speech and source separation, the four tasks LocalAI's AudioTransform can represent. AudioTransformResult carries one dst while htdemucs and mel_band_roformer produce several named stems from a single run, so inference runs ONCE, every stem is written as a sibling file <dst-stem>.<name>.<ext>, and params[stem] selects which one dst receives, defaulting to vocals and falling back to the first output. An unknown stem name is INVALID_ARGUMENT listing the real stem names rather than a silent substitution, and the selection happens before the first write so a refused request leaves no files behind. params[stem] is consumed here and is not forwarded into the engine's request options. The stem decision lives in stem_selection, which is stdlib only and therefore tested by backend/cpp/run-unit-tests.sh. It also validates the names, because they come from the model (htdemucs reads them from the GGUF's config.sources) and each becomes a component of a path this backend writes: a name carrying a path separator would escape the caller's output directory, and two stems sharing a name would silently overwrite one another. Both files are read at their native rate and channel count. Separation forces it, since demucs and roformer refuse any rate but 44.1 kHz and lose the stereo image that separates a centred vocal from a wide mix. The conversion families all resample internally (seed_vc, vevo2, miocodec, chatterbox were each checked), so passing the file through unchanged is also strictly better than band limiting it to 16 kHz first. Verified end to end against htdemucs f16 on a 44.1 kHz stereo mix: four stems plus dst, dst byte identical to the selected stem, params[stem] selecting a different one, an unknown stem refused with no files written, and mono input preserved as mono output. Also against miocodec for the single output path, where params[stem] is refused rather than ignored. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): refuse an impossible stem early, and stop blaming the caller for a failed write Four fixes from the first review of the AudioTransform RPC. check_can_serve now returns the resolved route, so params[stem] on a route that is not source separation is refused from the route instead of after a full inference: 11 ms rather than the 4.5 s a miocodec conversion costs, and far worse on seed_vc or vevo2. The post-run refusal stays as the backstop for a separation-routed family that returns no stems anyway. The typo'd-stem-name case still needs the run, since no framework header publishes the stem names before one. Stem names carrying control bytes are refused. GGUF strings are length prefixed and demucs reads its sources from JSON, so an embedded NUL survives to here: two names differing only after the NUL are distinct std::strings, so the duplicate check passes them, and then path::c_str() truncates both and they open the same file. That is exactly the silent overwrite the duplicate check exists to prevent, with the .wav lost as well. A failed write is now INTERNAL rather than INVALID_ARGUMENT. The destination is LocalAI's own generated-content directory, not anything the caller named, so a full disk or a permission fault there is a server fault and is worth retrying, which is the opposite of what INVALID_ARGUMENT tells a client. An empty output path stays INVALID_ARGUMENT. Two comment corrections and one clarification: the separators' required rate is their checkpoint's declared samplerate rather than a hardcoded 44100, seed_vc resamples with soxr and falls back to sinc-hann, and the "no files left behind" guarantee covers a refused request, not a write that fails partway through the loop. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(audio-transform): stop folding every upload to 16 kHz mono, and name the separation stems Two defects that made source separation unusable through LocalAI's own API, even though the backend served it correctly over gRPC. /audio/transform normalized every upload to 16 kHz mono s16 through utils.AudioToWav, with no way past it. htdemucs and mel_band_roformer refuse any rate but their checkpoint's own and separate a centred vocal from a wide mix using the stereo image, so every separation request through the HTTP API died with "HTDemucs prepare() sample rate mismatch: expected 44100, got 16000" while the same call over gRPC worked. The fold is not wrong, it is backend-specific: LocalVQE's echo cancellation genuinely wants 16 kHz mono and needs the reference in the same shape. So it becomes a declaration, BackendCapability.AudioTransformInputMono16k, set for localvqe and for nothing else. A backend that declares nothing gets its upload unchanged, which means no backend has to opt in to work. utils.AudioToWavPreservingShape is the non-folding conversion: a 16-bit PCM WAV passes through byte for byte at any rate and channel count, anything else is transcoded to WAV with its rate and channel layout kept. The other defect is that the run-once stem design bought nothing. A separation backend writes every stem beside dst from one inference, but AudioTransformResult carried only dst, so the other three were files no caller could find and a caller wanting all four had to run four separations. AudioTransformResult grows a repeated AudioTransformStem, the backend fills it, core/backend validates that each path really is inside the generated-content directory it handed over, and the endpoint publishes them as an X-Audio-Stems JSON header beside the existing X-Audio-Input-Url. JSON because a stem name is the model's own string and could contain any separator a hand-rolled format would use. Verified end to end through the HTTP endpoint with htdemucs f16 on a 44.1 kHz stereo file: 200 with a 44.1 kHz stereo body, all four stems named and fetchable through /generated-audio/, body byte identical to the selected stem, and params[stem]=drums returning a different one. The same upload sent to a model whose backend is localvqe still reaches the backend as 16 kHz mono, confirmed both by the engine's own rate refusal and by the persisted input file. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(audio-transform): reject extensible WAV from the passthrough, escape stem URLs, convert stems with dst Four fixes from the second review, plus one bug they made visible. isPCM16Wav tested only the bit depth, and go-audio's IsValidFile never looks at the format tag, so a 16-bit WAVE_FORMAT_EXTENSIBLE (0xFFFE) upload was passed through untouched where the old fold would have transcoded it. audio.cpp's WAV reader accepts 16-bit only when the tag is 1, so such a file died with "unsupported WAV encoding". Extensible is what many DAWs and Windows tools write and music files are this endpoint's new headline input, so it is a first-contact failure rather than a corner. The check now requires tag 1, with a spec that fails against the old implementation. Stem URLs are percent-escaped. A stem name is the model's own string and legally contains a space, a '#', a '?' or a '%'; an unescaped '#' truncates the URL before the request is even sent. The name field keeps the raw name. sample_rate and response_format are applied to the stems as well as to dst. Applying beat documenting: dst IS one of those stems, so leaving them alone broke the "dst duplicates the selected stem" invariant the whole design rests on, and both conversions are no-ops when unset. A stem whose conversion fails is dropped from the header rather than advertised in the wrong shape. Verifying that turned up why it had never been noticed: the two fields were never bound at all. The request arrives as multipart/form-data and echo's binder falls back to the FIELD NAME without a form tag, matching only case-insensitively, so "SampleRate" never matched "sample_rate" and "Format" never matched "response_format". Both were documented in the endpoint table and silently ignored. Two form tags fix it, and with them the conversion is observable end to end. Docs: audio-transform.md now documents what LocalAI does to an upload before the backend sees it, which backend gets the 16 kHz mono fold and why, params[stem], and the X-Audio-Stems header with a worked example. Also records the known limitation that the fold lookup is on the bare backend name, so pinned variants (vulkan-localvqe) do not match, and points at IsLlamaCppBackend as the suffix-tolerant precedent. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): serve the TTS and SoundGeneration RPCs TTSRequest.voice is treated as a speaker reference clip when it names an existing regular file, which makes routing prefer VoiceCloning, and as a named preset otherwise, in which case it travels as VoiceReference::cached_voice_id. Both the clip and SoundGenerationRequest.src are read at the file's own rate and channel count: upstream's own CLI and server do exactly that, every consuming family resamples internally and mostly with a better resampler than ours, and ace_step and stable_audio resample their input per channel, so a downmix here would delete the stereo image they are built to consume. The request builders live in their own unit rather than in grpc-server.cpp's anonymous namespace so they can be tested; grpc-server.cpp has a main() and cannot be linked into a test binary. The option keys are the whole point of these functions, so each one was grepped against the pinned upstream and the accounting is written down beside it. instructions maps to "instruct", which is what upstream's own server maps the OpenAI field to and what qwen3_tts and omnivoice read, and to "caption" for irodori_tts; the style tag is spelled "instruct" too, because "instructions" is looked up nowhere. duration maps to "duration_seconds", read by all three generation families, with the proto's own name kept only as a forward-tolerant alias. Keys that no family reads say so. Both handlers answer a capability refusal before taking the lane and before any file read, so a model that cannot synthesise does not queue behind somebody else's run to be told no. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): stop emitting an empty style language, and name the missing clip StyleCondition::language was set whenever has_language() was true, with no !empty() guard, while the language option twelve lines below had one. core/backend/tts.go sets Language unconditionally, so has_language() is true on every request LocalAI sends and carries "" when the caller named none. An engaged-but-empty style language is worse than an absent one: supertonic reads text_input->language behind its own !empty() guard and then overrides it from style->language with no guard at all, so "" replaced its "en" default and its tokenizer threw "invalid Supertonic language: ". Every /v1/audio/speech request that set instructions and no language would have been an INTERNAL against a supertonic model. A plain request never saw it, because the style condition only exists when instructions are non-empty, which is why the chatterbox end to end run did not catch it. TTS also stops discarding the Route that check_can_serve already returns. A family routed to voice cloning without a reference clip used to be refused from inside its own prepare(), which meant an INTERNAL naming neither the RPC nor the field to set; chatterbox advertises clon and no tts, so that was every preset-only request to it. It is now an INVALID_ARGUMENT naming TTSRequest.voice, answered in about 4 ms, and it cannot misfire because has_voice_reference is what selected cloning in the first place. Reading CapabilitySet::supports_speaker_reference to generalise this stays a follow-up. The src read carries a written caveat rather than a family blocklist, because ace_step's editing routes legitimately need src: setting src on a stable_audio model corrupts the heap and aborts the process in the pinned upstream, and the only thing keeping that off the network is that schema.ElevenLabsSoundGenerationRequest has no field for it. Nobody reading that Go schema would know why, so the reason is recorded where the field is read. build_tts_shape is extracted so TTSStream cannot describe the same request differently, and it arrived untested: two mutations of it survived until a test_tts_shape case was added. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): serve the TTSStream and AudioTranscriptionStream RPCs TTSStream leads with a streaming WAV header carrying 0xFFFFFFFF sizes, matching the convention backend/go/vibevoice-cpp established, so an HTTP client can start playback before the full PCM exists. Its chunks are read from StreamEvent::named_audio_outputs and not audio_output: supertonic, omnivoice and voxcpm2 all put their streamed audio there and leave audio_output empty until the very end, so reading the obvious field yields a stream with no audio in it. The finish_stream result is the family's own merged whole rather than a tail, so it is emitted only when nothing was streamed. Streaming transcription sends incremental deltas and degrades to a single delta plus the final result on families that offer no streaming ASR, which is the same message sequence with fewer deltas. The four streaming ASR families disagree on what partial_text means: nemotron_asr, vibevoice_asr and higgs_audio_stt report incremental fragments while voxtral_realtime reports the whole hypothesis and reports it twice, so the reconciliation lives in one tested unit rather than in the handler. nemotron_asr reports only through the stream event sink, and only from inside finalize, so the audio driver installs one and clears it again before returning: the session is cached and a sink left holding the caller's frame is a use after free waiting for the next stream. begin_stream is now the only implementation of the streaming state obligation, prepare then start_stream. Streaming sessions are cached, and what clears the previous stream is start_stream's reset; a family override that dropped it would break every call site with no compile error, so there is one call site. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): keep streaming deltas on UTF-8 boundaries, refuse dtypes that abort TranscriptStreamResponse.delta is a proto3 string, whose wire format requires valid UTF-8. voxtral_realtime reports its hypothesis as a concatenation of raw token BYTES (tokenizer_text.cpp:171-183), so the cumulative difference between two consecutive reports is eventually a lone continuation byte, and the C++ runtime serializes that with only a logged warning while the Go runtime refuses to unmarshal it: the client loses the remaining deltas AND the final_result. Measured on a trace of a non-ASCII sentence, 11 of 31 messages failed to unmarshal and every accented character was lost. TranscriptDeltaTracker now holds back an incomplete trailing sequence and merges it into the next fragment; reconcile flushes it, which it always can because the final text is complete. The same trace now unmarshals in full with zero failures. A streaming buffer whose float count is not a whole number of frames is refused rather than truncated. The integer division dropped the tail floats from the fed audio and therefore from the transcript, with no diagnostic; vibevoice_asr refuses the same thing from the other side of the call. A supertonic GGUF whose weights are not f32 is refused at load. It reaches ggml_concat with mismatched operand types and ggml_abort takes the whole backend process down on the first request, so nothing downstream can report it: the model loads, then every request kills the process. Attributed rather than assumed, the unary TTS path aborts identically, and upstream records that package as untested. The refusal names the orig package and says what to run before deleting the guard. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): stop a repeated lead byte from orphaning the next delta The first UTF-8 fix closed the cumulative half only. Rule 2 discards a fragment the known text already starts with, and when that fragment is the LEAD BYTE of a new character it looks exactly like a repeat of an older character beginning with the same byte. It was discarded rather than held, its continuation bytes then arrived alone and began the next delta, and utf8_complete_prefix_length only ever inspected the trailing sequence, so a delta invalid at the FRONT went out whole. Through a real Go proto.Unmarshal the review's four-character repro gave 3 deltas, 2 unmarshal failures and a lost transcript. Reachable from the incremental families, not only from voxtral: nemotron_asr's decoder cuts at a byte offset and vibevoice_asr's common_prefix_size compares bytes, so both split characters. Measured over 30,000 randomized incremental traces, 53.28% of Japanese traces and 9.52% of French ones carried at least one delta the Go runtime refuses. Two changes. Rule 2 no longer judges a fragment that ends mid-character, so the lead byte is held instead of swallowed and the character survives intact; the cost is a few duplicated bytes in a shrinking cumulative report, which no pinned family produces. release() additionally drops leading orphan continuation bytes, so no delta can begin mid-character whatever the rules above it decide. Losing a byte keeps the stream alive; emitting one ends the RPC and takes the final_result with it. Post-fix all 60,000 traces produce zero unmarshal failures, and the cumulative streams plus both pure-ASCII incremental streams are byte-identical to the previous commit, so nothing changed for the families already working. The weight-dtype allow list moves to family_gate, where it is stdlib-only and pinned by a test rather than only by a comment. Two comment citations corrected. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): read only an exact repeat as a repeat, not any prefix Rule 2 discarded any partial the known text merely started with. For a cumulative family that is a duplicate; for an incremental family it is an ordinary short fragment that happens to coincide with the start of the transcript, and it was dropped, silently corrupting the text. Pure ASCII, no multi-byte character anywhere: the fragments "pure ", "ascii ", "trans", "c", "ri", "p", "t" left the client holding "pure ascii transcrit". Over 5,000 randomized traces per transcript, 9.50% of pure-ASCII and 29.12% of French traces ended with the client holding something other than final_result.text, with a 200 and no diagnostic. Both incremental families emit fragments that small routinely, since nemotron_asr cuts at a byte offset and vibevoice_asr at a common prefix. Narrowing rule 2 to an exact repeat drives that to zero on all six transcripts and changes no cumulative stream at all: 30,000 randomized cumulative traces are byte-identical to the previous commit. What rule 2 guarded was established from upstream rather than from its own comment. The only duplicate any pinned family produces is voxtral_realtime's, where process_available_stream_chunks feeds each event to the sink from inside its loop and returns the last of the batch, so that event arrives twice with byte-equal text. A duplicate is an exact repeat, so equality still covers it. The case given up is a cumulative report that SHRINKS, which no pinned family can produce: voxtral decodes a token vector that is only push_back'ed and cleared by reset(), so within a stream it can only grow. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): serve the AudioTranscriptionLive RPC The one bidirectional stream this backend serves. The client sends a TranscriptLiveConfig, then TranscriptLiveAudio frames; the server acknowledges with ready, emits deltas as the audio arrives, and sends final_result once the read side closes. There is no offline fallback: live transcription has to consume audio incrementally, so a family with no streaming ASR is refused rather than served a batch run, which is what this RPC's Streaming-only mode_candidates list already says. The driver is a new sibling of run_streaming_audio, run_streaming_live, because the audio does not exist yet: instead of slicing a buffer it pulls frames from the caller until the read side closes. It installs the same ScopedStreamSink in the same order, which is not optional, since nemotron_asr returns a bare event from process_audio_chunk and reports every partial through the sink from inside finalize(). It buffers the wire's frames up to the family's own preferred window rather than feeding whatever size the client's audio callback produced, and it does not call finish_stream at all when no audio arrived, because nemotron_asr throws "finalize requires streamed audio" and an empty transcript is the truthful answer to transcribing nothing. Three things the handler had to get right and one it cannot: - The audio contract. A live request carries no samples, but nemotron_asr's streaming prepare() throws without an audio contract, and build_preparation_request derives it from TaskRequest::audio_input, so that field is an EMPTY buffer holding only the rate and the channel count. - 16 kHz or a refusal. The families express their spans in their own 16 kHz feature domain whatever the input was, and live frames cannot be resampled on the way in the way a file can, so an 8 kHz session would return timestamps 2x off with a 200. core/backend hardcodes 16000 anyway. - A mid-stream Config is refused. backend.proto calls it a decoder reset, but deltas already on the wire cannot be retracted, so a reset would leave the final text contradicting the transcript the client assembled. Ignoring the message would hand a client that believes it reset the decoder a transcript that silently continues the audio it thought it discarded. - The stale-route identity check cannot run here: TranscriptLiveRequest carries no ModelIdentity in either arm of its oneof, so snapshot_for does not instantiate for it. snapshot_unchecked's comment now names that as a second legitimate class of caller and says the fix is a proto change. eou and eob stay false. They exist for cache-aware models that emit end-of-utterance and end-of-backchannel tokens; audio.cpp's StreamEvent has no equivalent signal, and a client uses eou to decide the speaker yielded the turn, so a guess inferred from silence cuts people off mid-sentence. The lane is held for the whole stream, which is as long as the user keeps talking: the streaming session is stateful and cached, so a concurrent run would interleave two callers' audio and corrupt both transcripts. Verified against nemotron_asr over a real connection with a 14 s WAV in 512-sample frames: ready first, 59 incremental deltas with no repeated prefix, concat(deltas) equal to final_result.text, word timestamps in nanoseconds, eou and eob false. citrinet_asr answers UNIMPLEMENTED naming the family and listing asr/offline. A config followed by a close returns an empty final_result rather than hanging, and a first message that is not a config is INVALID_ARGUMENT. Two concurrent streams both return the complete transcript. Two cleanups on lines Task 12 touched, folded in. The DtypeAllowList terminator is now asserted at compile time: the reported out-of-bounds read did not exist, the single entry does terminate, but the loops have no other bound and any edit that widened an entry would walk off the end. And the dtype guard now short-circuits on "is there a table entry" through a new predicate rather than on the emptiness of the description string, which would have skipped the check on an entry with an empty allow list, i.e. on precisely the entry that refuses every dtype. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): bound the lane a live stream can hold AudioTranscriptionLive holds the model's inference lane for the whole stream, which is correct (the streaming session is stateful and a concurrent run would interleave two callers' audio) and newly dangerous. Every other RPC holds the lane across compute, or across a write to a slow reader, and both of those terminate on their own. A live stream instead blocks in a client-driven read, and a peer that goes silent WITHOUT closing the stream never terminates anything: the lane stays taken and every other request against that model queues behind a client that stopped speaking. live_watchdog is a one-shot idle timer that ends the stream when no frame has arrived inside a window. It is standard library only, so it is unit tested without an engine. gRPC's synchronous Read has no timeout and cannot be given one, so the only way to unblock it is ServerContext::TryCancel, which decides the wire status itself: the client sees CANCELLED rather than the DEADLINE_EXCEEDED the handler returns, the reason is logged, and the lane coming back is the point. When it fires the read loop throws rather than reporting end-of-input, so the driver does not go on to finalize a decode nobody is waiting for. It is armed only after the lane is taken and disarmed as soon as the read side closes, and both ends matter. Arming earlier would cover acquire(), which legitimately blocks while another live stream runs, so a queued caller would be cancelled for waiting its turn. Disarming later would cover our own decode, where a window overrun is not a peer going quiet and cancelling would throw away the transcript the client is waiting for. The window is the new live_idle_timeout_ms option, 30 s by default, 0 meaning no limit. core/http/endpoints/openai/realtime.go drives a 300 ms ticker and feeds every tick that produced new audio while a turn is open, so 30 s of silence is a hundred ticks that delivered nothing. It is also longer than any pause a speaker takes mid-utterance, which is the case that must never be cut off, and backend.proto lets one stream span many utterances, so a client that pauses longer between them raises the option rather than discovering it. Two smaller corrections in the same handler: - check_can_serve now runs BEFORE the sample rate check. pkg/grpc/grpcerrors/errors.go degrades to the file path on UNIMPLEMENTED and on nothing else, so a live-incapable model asked at a wrong rate was answering INVALID_ARGUMENT and costing the caller its fallback. - a negative sample rate is refused instead of silently becoming 16000. Zero still means 16000, which is what the proto documents; -1 is malformed rather than absent and gets the same refusal every other bad rate gets. And one thing recorded rather than changed, at the handler: "live" here means incremental INPUT, not low latency, and with the pinned families it does not yet mean incremental OUTPUT either. nemotron_asr's process_audio_chunk only appends to its buffer, so its whole decode and every delta happen inside finalize(), after the client closes its send side. The policy-window buffering is inert for that family and matters only for vibevoice_asr and higgs_audio_stt. Verified on the wire with live_idle_timeout_ms:3000. A silent client acked at 371 ms and was cancelled at 3.371 s; a second live stream opened one second later received its ack 2.37 s in, i.e. at the instant the first was cancelled, and then transcribed successfully on the same cached session. Without the watchdog it would still be waiting. Re-ran the live transcription (ready first, 59 incremental deltas, concat equal to the final text, word timestamps in nanoseconds, eou and eob false), the citrinet refusal at both a right and a wrong rate (UNIMPLEMENTED either way now), and Task 12's AudioTranscriptionStream on nemotron_asr, which is unchanged. Mutation testing the watchdog found a weakness in its own test: the destructor test slept past the window inside the watched scope, so a destructor that DETACHED the thread instead of joining it passed unnoticed. The test now uses a window longer than the scope, which kills that mutant, and says why. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): refuse the unsupported RPCs with a reason AudioEncode, AudioDecode, AudioTransformStream, AudioToAudioStream and VoiceEmbed have no counterpart in audio.cpp's VoiceTaskKind. Each now returns UNIMPLEMENTED naming the loaded family, what that family does support, and the upstream limitation, instead of the generated base class's bare status. The reasons live in a table in capability_routing.cpp so they are data rather than literals copied into five handlers, and so a test can assert every one of them. The five claims this was planned against were re-read at the pinned upstream e800d435d130dc776baf6f3e6129bb62b1495c89, and one did not hold. "audio.cpp streams tts and asr only" is false: silero_vad advertises vad with RunMode::Streaming. The refusal stands on the narrower claim that survives, that no family advertises streaming for any task AudioTransform routes to, and a test asserts the refuted wording does not come back. VoiceEmbed is the one refusal whose request carries a ModelIdentity, so it runs the #10952 check before answering: a stale route must get NOT_FOUND and the router's sentinel, not "audio.cpp cannot embed speakers" about a model that is not loaded here. It cannot use snapshot_for, whose no-model branch would tell the caller to load a model when no model can help, so it takes the reference through snapshot_unchecked and checks identity itself. That function's comment now names three classes of caller instead of two. The two bidirectional surfaces refuse without reading their stream, verified with a client that writes a config and eight frames first and gets the status rather than hanging. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): correct the vevo2 clause, and assert the absences Review found a false clause in the AudioToAudioStream refusal. It said s2s is "offline voice conversion ... which converts one clip into another speaker's voice", which is true of miocodec and false of vevo2: vevo2's s2s route is `editing` and only `editing` (default_route_for_task and route_matches_task in src/models/vevo2/session.cpp), documented as "Edit source speech into new target text while using the target voice" and requiring --target-text, so it rewrites what was said. vevo2's voice conversion is its separate vc task. It now reads "offline clip-to-clip processing against a target voice, declared only by miocodec (voice conversion) and vevo2 (speech editing)", and a test asserts the miscast cannot come back. The conclusion is unchanged: neither family converses. That defect was undetectable on the wire, since vevo2 does not load here, which is the argument for upstream_absence_ctest.cpp. It links engine_runtime purely to interrogate make_default_registry() and asserts the five premises the refusal reasons rest on: no codec task kind, no family advertising spk, no streaming for sep/vc/svc/s2s, miocodec advertising exactly vc and s2s, and s2s advertised by exactly miocodec and vevo2. The last two are exact sets, so an addition fails here rather than leaving a message stale. A positive control proves the registry is populated and the query works before any absence is believed, and every assertion has a reproduced negative control. This turns an AUDIO_CPP_VERSION bump from "remember to re-read five prose paragraphs" into a test failure. unsupported_surface now switches over UnsupportedRpc with no default label, so -Wswitch reports a sixth enumerator added without a row at build time; the runtime bounds guard it replaces is deleted. The AudioTransformStream reason had a true premise and an overreaching conclusion: an offline sep family could be buffered into a stream, as other LocalAI backends do. It now says this backend declines to offer a buffered offline call in disguise, rather than implying impossibility. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): make the missing-switch-case diagnostic fatal unsupported_surface() switches UnsupportedRpc onto the table row that explains it, with no default label, so -Wswitch reports an enumerator nobody handled. As a warning that is not enough: adding a sixth enumerator and building the shipping target gives exit 0, a binary and one warning, and the trailing `return surfaces[0];` then answers the new RPC with AudioEncode's codec reason. That is a confident, specific and false statement about audio.cpp on the wire, on the one code path whose entire job is to be truthful about what this backend cannot do, and it is worse than the runtime fallback it replaced, which at least named itself as a bug in this file. capability_routing.cpp therefore joins loaded_model.cpp on the existing -Werror=switch pin, whose comment already made this argument for the engine enum. The comment now covers both files. The pin stays per-file rather than project-wide because upstream's own ace_step/vae_decoder.cpp has unhandled -Wswitch cases of its own. Verified: a sixth enumerator now fails `make grpc-server` with exit 2 and no binary; appending a 14th VoiceTaskKind upstream still fails loaded_model.cpp, so the two pins fire independently; both reverted clean. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): package the backend image Bundles the dependency closure for the from-scratch image, the dlopened ggml CPU-variant shared objects that ldd cannot see, and upstream's bundled silero_vad and marblenet_vad assets so VAD works with no download. The bundled loader sits in the package ROOT rather than at lib/ld.so. run.sh execs it, which makes /proc/self/exe name the loader, and this backend has two consumers of that path: ggml discovers the libggml-cpu-*.so by listing dirname(/proc/self/exe), and resolve_model_path expands bundled:<name> under the same directory. Rooting the loader makes the binary, the ggml objects and assets/ share the one directory all three resolution mechanisms agree on. llama-cpp's lib/ld.so layout would need assets/ moved into lib/ as well. The image builds against apt gRPC and protobuf, like Dockerfile.ds4 and unlike Dockerfile.privacy-filter. The from-source gRPC that install-base-deps.sh and the base-grpc-* images supply vendors protobuf 26, which pulls abseil into message_lite.h; with SPM_PROTOBUF_PROVIDER=package that collides with sentencepiece's vendored mini-abseil and every absl::internal reference becomes ambiguous. Noble's protobuf 3.21.12 predates the abseil dependency and is the pair every earlier verification of this backend ran against. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): exempt the driver libraries from the packaging gate package.sh already left libcuda.so* and libnvidia-* to the host when copying, because the driver has to match the kernel module on whatever host runs the image, but the validation gate had no matching exemption. With BUILD_TYPE=cublas ggml is static and links CUDA::cuda_driver, so grpc-server carries DT_NEEDED libcuda.so.1 and the gate would have rejected the very absence the copy loop created, failing every cublas build in CI. One regex now feeds both. Building a control for that found a second defect: ld.so --list refuses to trace an object with an unresolvable dependency at all, exiting 127 without emitting a per-library line, so the "=> not found" rule was dead code and no exemption could have applied to it. The gate now traces with LD_TRACE_LOADED_OBJECTS and LD_LIBRARY_PATH, which reports the missing name and exits 0, and which is also what run.sh does at run time. Adds a layout assertion so a future move of the loader into lib/ fails the build instead of shipping a package that resolves bundled: models into lib/assets and finds no ggml CPU backend, and records for Task 16 that the Darwin script must not be a straight copy of privacy-filter-darwin.sh, which never calls package.sh and would silently drop assets/. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): register the backend with CI and the gallery Adds the five Linux matrix entries (cpu amd64/arm64 sharing a tag-suffix so the manifest merge fires, cuda 12, cuda 13, vulkan), the path-filter case that keeps later PRs touching backend/cpp/audio-cpp/ from getting zero CI jobs, the bump-bot entry pointing at the AUDIO_CPP_VERSION pin in the backend Makefile, the gallery meta plus its -development variant and the image entries for every variant, and the Makefile docker-build wiring. The matrix entries carry base-image only, with no builder-base-image, unlike the llama-cpp and privacy-filter blocks they sit next to. The prebuilt quay.io/go-skynet/ci-cache:base-grpc-* images ship a from-source gRPC whose protobuf v26 depends on abseil, and this backend's sentencepiece is built with SPM_PROTOBUF_PROVIDER=package, so it sees real abseil's absl::lts_20240116:: internal alongside its own vendored plain absl::internal and every absl::internal:: reference becomes ambiguous. Building against base-grpc-amd64 fails at sentencepiece-static.dir/error.cc.o with "reference to 'internal' is ambiguous". Dockerfile.audio-cpp installs apt's gRPC/protobuf 3.21.12 itself, which is also the pair every unit and end-to-end run of this backend has been verified against, and the CUDA toolkit therefore has to come from base-image. No Darwin matrix entry and no metal gallery entries: the Metal build needs scripts/build/audio-cpp-darwin.sh, a backends/audio-cpp-darwin make target and a routing step in backend_build_darwin.yml, none of which exist yet, so an entry added now would be routed to build-darwin-go-backend and look for backend/go/audio-cpp/. The inferBackendPathDarwin case and the DARWIN_BESPOKE_BUILDERS membership are in place, inert, so that adding the entry later is a one-line change that cannot be claimed by the generic Go path. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): pin the CUDA architectures, drop the vulkan variant Upstream sets CUDA_ARCHITECTURES to `native` on the engine_runtime target whenever CMAKE_CUDA_ARCHITECTURES is unset at root scope, and docs/build/ linux.md says so outright. ggml's own default does not rescue it: it list(APPEND)s in the ggml subdirectory scope, which never reaches the root scope where the engine_runtime property is decided. No CI runner has a GPU for `native` to enumerate, so both cublas entries would have gone red on the very commit that first turns a CUDA build on. Pin the list in backend/cpp/audio-cpp/Makefile, selected by CUDA_MAJOR_VERSION, which Dockerfile.audio-cpp now forwards from the CI build-arg it was previously discarding. The values are copied from ggml's own version guards rather than invented, so engine_runtime and ggml compile for the same set: CUDA 12 keeps the Maxwell/Pascal/Volta virtual archs and stops at 120a-real, CUDA 13 drops them and adds 121a-real. The `a` suffix is used rather than `f` because the latter needs CMake 3.31.8 and Ubuntu Noble ships 3.28.3. Verified by driving CMake 3.28.3's own CUDA architecture validator over both lists, with 120f-virtual as the rejected control. Drop the vulkan matrix entry, its two gallery entries, the vulkan capability key on both metas and the Vulkan tag. Every other vulkan backend gets its Mesa ICD drivers from .docker/install-base-deps.sh, which package-gpu-libs.sh then bundles; Dockerfile.audio-cpp calls neither and installs only libvulkan-dev and glslc, so the image would ship a Vulkan loader that finds no GPU. No CI job runs a vulkan image against real hardware, so that would have passed green and failed in users' hands. BUILD_TYPE=vulkan stays supported for local builds. Also note on the cublas entries that cuda-major-version now selects the architecture list and that cuda-minor-version and the base-image tag encode the same toolkit, and correct the stale entry counts on matrixEntryKey. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): build for Darwin Metal Bespoke C++ Darwin path like ds4 and privacy-filter: an includeDarwin matrix entry, a backends/audio-cpp-darwin make target, a gated workflow step, and the metal image entries plus metal/metal-darwin-arm64 capability keys in the backend gallery. The build script deliberately does NOT reassemble the package the way privacy-filter-darwin.sh does. It runs the backend's own `make package` and copies the result, so the Darwin package keeps the root-level layout the Linux one has: grpc-server, run.sh, the ggml objects and assets/ in one directory, with lib/ for the dylib closure. Hand-assembling would drop assets/, and assets/ is what makes the bundled: model paths resolve with nothing downloaded. The dylib walk is a full transitive closure rather than the single level ds4 and llama-cpp do, because Homebrew's grpc++ pulls libgrpc, abseil, upb, cares and OpenSSL that grpc-server does not link itself, and a level-1 walk ships a package that only works on a machine that already has Homebrew grpc. Two fixes folded in, both in the backend Makefile: - an EMPTY CUDA_MAJOR_VERSION fell through to the CUDA 12 architecture list, which contains 120a-real and so needs nvcc >= 12.8. A local BUILD_TYPE=cublas build on a 12.0-12.7 host failed to compile where upstream's documented default (native) worked. EMPTY now maps to native, 12 and 13 keep their lists, and any other non-empty value is an error on cublas builds. CI always passes a major, so CI is unaffected. - the Darwin branch now points CMake at Homebrew's keg-only libomp. AppleClang ships no OpenMP runtime and nothing is symlinked into /opt/homebrew, so FindOpenMP finds neither the library nor the header, and audio.cpp calls find_package(OpenMP REQUIRED) whenever ENGINE_ENABLE_OPENMP is on. Without the hint the macOS build would have died at configure time. If the keg is absent the build disables OpenMP instead of failing. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): make the Darwin fallbacks loud and the rpath walk complete Review follow-up on the Darwin Metal build. The OpenMP fallback was silent. If brew --prefix libomp ever comes back empty, CI produced a green Metal package with 108 #pragma omp directives across ~30 files compiled out, and clang says nothing about an ignored omp pragma without -Wsource-uses-openmp, so the only trace was one absent flag inside a set -x cmake line. That regression would have been blamed on Metal. It now warns. The @rpath arm of the dylib walk had no live candidate when it was written, on the reasoning that a Metal build links ggml statically. The OpenMP fix in the same commit made libomp.dylib one, and whether Homebrew records it as an absolute opt path or as @rpath/libomp.dylib is not observable from Linux. The walk now expands @rpath, @loader_path and @executable_path against the object's own LC_RPATH entries, and only fails when nothing on disk answers, printing the rpath list with the error so a failure on a machine nobody can attach to explains itself. Also: ADDITIONAL_LIBS now go through the closure rather than a bare cp, so they are deduplicated and their own dependencies bundled; build/darwin/lib is created explicitly instead of relying on package.sh pre-creating it; the libomp probe uses nested ifneq rather than $(and ...), which needs GNU make 3.81 and would otherwise expand empty and take the OFF branch on an older make; and -DOpenMP_ROOT is quoted like its CUDA sibling. Verified with a Linux harness that runs the script verbatim against a stubbed otool: a level-2 transitive dep, an @rpath dep reachable only through LC_RPATH, and an ADDITIONAL_LIBS dep are all bundled, a dependency cycle terminates, system libraries are skipped, the packaged tree has assets/ at the root beside grpc-server with the dylibs in lib/, and both failure paths exit non-zero. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): make bundled: reachable from a model YAML resolve_model_path() tested the bundled: prefix on `candidate`, which prefers ModelFile and falls back to Model. LocalAI fills ModelFile by joining ModelPath onto the configured model string (pkg/model/loader.go, LoadModelWithFile), and only sets it from a managed artifact otherwise, so a model YAML saying `model: bundled:silero_vad` arrives as ModelFile "/models/bundled:silero_vad" and Model "bundled:silero_vad". The prefix therefore never matched through the normal load path: it matched only for a hand-written LoadModel call that left ModelFile empty, which is exactly how task 15 verified it, and every model YAML using the form failed with "model path does not exist: /models/bundled:silero_vad". Both fields are now checked, Model first, so the zero-download VAD path the package ships assets for is reachable the way it is documented. A caller that puts the form in ModelFile still works, so task 15's verification stands. Compiled clean; the runtime check could not run on this host, whose system libprotobuf/libre2 have gone missing (the pre-existing grpc-server binary no longer resolves its libraries either), so it wants a container run. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): advertise the backend and document its options Registers audio-cpp as preference-only in /backends/known: the family lives in GGUF metadata that an importer cannot read from a remote repo, and one repo hosts thirty families, so there is no honest auto-detect signal. Modality is a single string and the import form chips on a fixed key set, so it registers as tts with the other modalities named in the description rather than under an invented key the UI would bucket as "other". Adds a features page covering the option namespacing, the routing table per endpoint, the RPCs this backend declines and why, the bundled VAD path, the separation stem behaviour, and the family gotchas (supertonic needs the orig package; chatterbox advertises cloning and no plain tts; nemotron_asr defers its whole decode to finalize so live transcription emits nothing until the client half-closes, unlike higgs_audio_stt and voxtral_realtime). Every option name and family capability in it was read off the pinned upstream checkout. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * backend(audio-cpp): test resolve_model_path, and correct the family names The bundled: fix in 842443cd7 shipped without a test, which is how the bug got there: task 15 verified the form with a hand-written LoadModel that left ModelFile empty, and that is the one shape the server never produces. Four cases in streaming_driver_ctest, which already links loaded_model.cpp, pin the PRODUCTION shapes instead. The first fails against the pre-fix source (returns the joined /models/bundled:silero_vad); the other three are the branches the bundled: lookup now runs in front of and must fall through for. Three family names in the docs were the source directory rather than the registered family, on pages whose whole argument is that these names cannot be guessed: demucs is htdemucs (demucs/loader.cpp:22), roformer is mel_band_roformer (roformer/assets.h:15), and moss is TWO families, moss_tts_local and moss_tts_nano. The hyphenated ASR names are underscored to match, here and in the compatibility table. The supertonic dtype note claimed more than the evidence carries. The f16 abort is a local observation, identical through TTS and TTSStream; upstream's docs/gguf.md leaves the 16-bit column untested and records q8_0 as "No (unsupported weight dtype)", which says unusable rather than fatal. Both are still refused, because the allow list is what the family can run. Corrected in family_gate.h, family_gate.cpp and the docs together, since the docs inherited the wording from the code. The importers tripwire says in the file that it is a tripwire: it exercises no audio-cpp behaviour, and the registration assertion lives in backend_test.go. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * gallery: add audio.cpp models covering every served RPC One representative model per RPC group of the audio-cpp backend, plus the two bundled VAD models, which need no download at all because the assets ship inside the backend package. Every hash was computed with sha256sum on the downloaded file. Quantizations come from upstream's tested-status table in docs/gguf.md rather than a default of q8_0: supertonic ships the orig package (its q8_0 is recorded as an unsupported weight dtype and its f16 aborts in ggml_concat), and nemotron_asr and htdemucs ship f16 because their q8_0 builds are recorded with drift while 16-bit is a clean pass. Diarization and separation use the diarization and audio_transform usecases, not transcript: /v1/audio/diarization and /audio/transform filter the default model on FLAG_DIARIZATION and FLAG_AUDIO_TRANSFORM respectively, so a transcript flag would have hidden both models from their own endpoints. The forced aligner sets parameters.language, which the transcription endpoint uses as the fallback when no language form field is sent, because the family requires both a transcript and a language. All ten entries were run twice: once against the raw gRPC server, and once installed with local-ai models install and called through the HTTP endpoint. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * gallery: correct the audio.cpp entries' licenses Swept all ten entries against the real upstream named in audio.cpp's tools/model_manager.py rather than against the audio.cpp repo's own license. Three were wrong: supertonic apache-2.0 -> openrail weights come from mlx-community/supertonic-3-mlx, and both it and Supertone/supertonic are openrail citrinet apache-2.0 -> other pulled from NGC nvidia/nemo/stt_en_citrinet_256, governed by the NGC Terms of Use sortformer other -> cc-by-nc-4.0 nvidia/diar_sortformer_4spk-v1 is CC BY-NC 4.0, and the gallery already uses that exact string, so there is no reason to obscure a non-commercial bar The license field is one word, so citrinet and sortformer also gained a sentence saying why they are restricted. The other seven were confirmed correct against their sources. Also drops an unverified claim from the nemotron description. It said the model drives the realtime transcription session; that endpoint actually calls TranscribeStream, and the live RPC reaches LocalAI only through realtime_semantic_vad.go. Neither path was exercised here, so the description now states only the two calls that were. MarbleNet gains the NeMo upstream under urls: for parity with silero. No sha256, quantization, usecase or model choice changed. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(audio-transform): bound sample_rate, keep same-named uploads apart Four defects the whole-branch review found on the Go side, plus two comment corrections. sample_rate is a disk-exhaustion hazard. The branch added the `form:` tag that makes the field bind for the first time, so the resample path went from dead to live, and utils.AudioResample interpolates the int straight into ffmpeg's -ar with no bound. Measured with ffmpeg 7: -ar 999999999 on a 0.01 s clip writes 20 MB and exits 0, which scales linearly to the reported 3.9 GB for one second, into a GeneratedContentDir nothing sweeps, and convertStems repeats it once per separation stem. Clamped to 8000..192000 in the handler, before the temp dir and before the model is touched, and rejected with a 400 outside it. The low end was reported as "a 0-byte file". It is not: -ar 1 writes a 78-byte header with no audio behind it, whose declared data size still claims 70 bytes, so go-audio parses it as a 35 SECOND file and a size check does not see it. The guard therefore compares the declared data chunk against the bytes actually on disk, and AudioResample now fails rather than returning a WAV carrying nothing. Both parts of a transform request land in one temp dir, and the raw copy was named only after the client's basename, so `-F audio=@mic/clip.wav -F reference=@loopback/clip.wav` wrote "raw-clip.wav" twice. Since AudioToWavPreservingShape hardlinks an already-PCM16 WAV rather than copying it, the reference part's os.Create truncated the inode audio.wav pointed at: mic and reference came out identical, which makes an echo canceller null everything and return near-silence with a 200. The raw copy now carries the form field name. audio-cpp had no BackendCapabilities entry, so VoiceCloningForModel returned nil before it ever consulted the model's tts.voice_cloning override and every `voice: "profile:<id>"` request was refused with a 400, on a backend that ships audio-cpp-chatterbox whose family serves cloning and not plain TTS. Registered with its RPCs, usecases and the reference-audio contract, and deliberately without the 16 kHz mono fold, which its separation families cannot survive. GetBackendCapability was exact-match only, so every pinned gallery variant read as an unknown backend: vulkan-localvqe lost the 16 kHz mono fold that used to be unconditional and started failing inside LocalVQE, and the usecase gate does not stand in for it because BuildFilteredFirstAvailableDefaultModel returns early once the client names a model. Lookup now falls back to the meta name by stripping the gallery's hardware prefix and release-channel suffix, exact match first so nothing can be shadowed. Same class as #10945. Also corrected: the AudioTransformRequest comment claimed echo's binder falls back to the field name, which it does not in either direction (bindData binds ONLY tagged fields and `continue`s otherwise; `model` arrives from setModelNameFromRequest's c.FormValue). And the stable_audio `src` heap corruption caveat now lives on ElevenLabsSoundGenerationRequest, where the Go developer who would add the field can see it, instead of only in C++. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(audio-cpp): refuse a task pin the RPC cannot serve, and stop empty frames holding the lane The model's `task:` option is copied into the request shape by all nine handlers, which is correct, but resolve_route then replaced the RPC's candidate list with the pin WHOLESALE and never asked whether the pin was something that RPC routes to. One pin therefore bled across all nine surfaces, and because the family still supported the pinned task the result was a wrong 200 rather than an error. Reproduced live: nemotron with task:asr made Vad return 200 with zero segments after a full ASR decode, so 14 seconds of speech was reported as silence, and Diarize did the same; silero_vad with task:vad made AudioTranscription return 200 with empty text and four segments whose spans were VAD segments, which combined with response_format in {text,srt,vtt,lrc} building the body solely from Segments[].Text yields a well formed SRT of four timed EMPTY cues. It also contradicted the documented contract, that a family which cannot serve a request is refused rather than rerouted. A pin is now checked against the RPC's admissible task set before it is adopted, and the refusal names both the pin and the RPC. The set is derived from task_candidates with every shape flag set rather than restated, so a task added to an RPC's candidates cannot become inadmissible by omission. Every legitimate pin survives, and the test asserts all fifteen of them alongside the eight crossings that must not. The live watchdog was defeated by empty frames. idle.touch() ran on ANY message, before the has_audio and pcm.empty() filters, so a peer writing unset-oneof or zero-length frames faster than the window held the lane indefinitely while feeding the decoder nothing. There is one lane per model and one model per process, so that is a single client denying the whole backend, which is what the watchdog exists to prevent, and the thrown text already said "no audio frame arrived". The touch moved below the filters, which are now a named predicate so the distinction is testable rather than a call order nobody can see. Three comments corrected against measurement rather than reasoning: - CMakeLists claimed zero google::protobuf:: definitions remain in the executable. nm -C --defined-only reports 2515, and that is expected: they are generated code, sentencepiece::ModelProto's own _InternalParse among them. The claim that holds, and the one the ABI fix is actually about, is that no vendored protobuf RUNTIME is linked and ParseContext::ParseMessage is UNDEFINED in the executable, resolving to libprotobuf.so. - refuse_cloning_without_a_clip's "cannot misfire" paragraph had its reasoning backwards. Routing picks VoiceCloning as the FALLBACK when there is no clip, which is the case being caught; chatterbox, which ships in the gallery, advertises clon and no tts at all, so every voice-less request lands there. - audio_units read "2.1 min at 96 kHz" for index 11289602, which is 1.96 min. 2.1 min is 96 kHz's OWN first failure at 12288002. Both were remeasured and the note is now a per-rate table. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * build(audio-cpp): exclude the upstream checkout from the C++ gate, harden the darwin walk run-unit-tests.sh pruned */llama.cpp/* but not */audio.cpp/*. It is safe today only by luck: upstream's 44 tests all put "test" at the FRONT of the filename (17 test-*.cpp, 27 test_*.cpp, zero *_test.cpp), so the glob misses every one of them, and nothing enforces that. This gate runs on every PR for every backend and compiles each match as a standalone translation unit with nothing but nlohmann/json on the include path, so the day upstream adds or renames one test the gate goes red repo-wide on an Apache-2.0 file nobody here wrote. audio-cpp-darwin.sh now logs the raw otool -L output and the parsed LC_RPATH list unconditionally, before the walk. Both awk filters in that script assume a column layout nobody working on this can observe, since it runs only on the CI Mac, and a green first Darwin run proves nothing about the assumption: an awk that silently matched nothing yields an empty dependency list, which reads exactly like "no non-system dependencies" and packages happily. Both filters otherwise feed process substitutions, so their input never reached the log. It also lists every symlink in the package and fails on one that cannot resolve inside the image. A dangling link does not fail anything else here, because every assertion tests with -e, which follows links; it fails at dlopen on a user's Mac. Links are NOT banned outright, which the review suggested but which would break the libggml.dylib -> libggml.0.dylib chain the `cp -a` above exists to preserve. What is banned is a link that resolves on the build host and will not resolve in the image: a broken one, or an absolute one pointing outside the package. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * style(audio-cpp): drop em dashes from the audio-cpp capability entry Follow-up to a84b3c4b9, no behaviour change. Assisted-by: Claude:claude-opus-5 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(config): key the voice-cloning model rule on the resolved backend Making GetBackendCapability strip the gallery hardware prefix and release channel fixed pinned variants of /audio/transform, but VoiceCloningForModel kept keying its per-backend switch on the caller's spelling. A pinned name therefore resolved the capability by stripping and then missed every case in the switch, falling through to the permissive default: cuda12-vibevoice-cpp advertised voice cloning for the realtime 0.5B model, metal-coqui for tacotron2, cuda12-crispasr for a pure ASR model, cpu-qwen3-tts-cpp for CustomVoice. Each of those is a model that cannot clone, so /v1/audio/speech accepted a profile: voice it had to fail on inside the backend rather than rejecting it with a 400, and the UI advertised the capability too. resolveBackendCapability now returns the key the entry was found under, and callers that branch on backend identity use that key instead of the name they were handed. The exact-match-first order is unchanged, so a backend genuinely registered under a variant-looking name still keys on its own name. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-5 [Claude Code] * gallery(audio-cpp): declare audio_transform on the chatterbox entry Chatterbox advertises VoiceCloning AND VoiceConversion (src/models/chatterbox), and the entry's own description already said so, but known_usecases listed only tts. /audio/transform selects its default model by FLAG_AUDIO_TRANSFORM, so voice conversion was reachable only by naming the model explicitly and was invisible to every usecase-driven surface. It is the one audio.cpp task with a shipped gallery model and no way to find it. Verified against the real model rather than inferred from the capability list: AudioTransform with chatterbox-q8_0, speech as audio_path and a speaker clip as reference_path, returns a 5.08 s 24 kHz mono WAV at -25.5 dB mean and zero stems, which is the single-output shape voice conversion should have. The description now says which endpoint reaches that half and warns that installing this next to a source-separation model gives /audio/transform two candidates, so the model should be named rather than defaulted. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-5 [Claude Code] * gallery(audio-cpp): add voice-design and singing-voice-conversion entries Two of the three audio.cpp task kinds that had no gallery model now have one. Both were driven end to end against the real weights through the backend before being written, not inferred from the capability tables. audio-cpp-irodori-voicedesign covers vdes. TTS carrying `instructions` routes to the vdes task, so the voice is described in words rather than supplied as a clip. Verified: "a calm elderly woman speaking slowly with a warm, gentle tone" over an 8.76 s 48 kHz mono render at -16.8 dB mean, and a closed-loop citrinet pass recovers the sentence with the accent drift expected from a Japanese-first model read by an English recogniser. audio-cpp-seedvc-singing covers svc, and pins task:svc because nothing else can reach it. seed_vc advertises svc and ordinary voice conversion, no request signal means "this input is singing", and auto-routing resolves the tie to voice conversion every time. Verified with the pin: 5.04 s 44.1 kHz output whose closed-loop citrinet transcription is exact. s2s deliberately has no entry, and the reason is not effort. miocodec is the only upstream family whose speech-to-speech route needs no text, and it returned audio with correct duration and level but no recoverable speech in four independent attempts: the stale build, v2 q8_0, v2 orig (the variant upstream records as a clean Pass), both tasks, and matched 44.1 kHz inputs on both sides. vevo2's route refuses with "Vevo2 text/prosody route requires text_input or target_text", and session.cpp:897 fills target_text only from request.text_input, which AudioTransform has no field to carry. The same vevo2 weights convert voice correctly through the default route with an exact ASR round trip, so the model and the plumbing are both healthy; it is the s2s route specifically that this RPC cannot express. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-5 [Claude Code] * backend(audio-cpp): carry transform text through params, add the s2s entry AudioTransform is audio-in / audio-out and its proto message has no text field, but not every task it routes to is audio-only. vevo2's speech-to-speech route is a text and prosody route: session.cpp:897 fills refs.target_text from request.text_input and nowhere else, and the run refuses without one with "Vevo2 text/prosody route requires text_input or target_text". The params map is the only channel this RPC has that reaches the engine, so the text travels through it and apply_transform_text_input unpacks it after the params have been copied into task.options. Before this, s2s was not awkward to reach through /audio/transform, it was unreachable, and it was the last audio.cpp task kind with a real model and no way to get to it. target_text is canonical and text is its alias, the order vevo2's own option table declares them in, so a request setting both gets the canonical one rather than whichever the map happened to store first. An empty value falls through to the next candidate instead of ending the search. language rides along only when a text was found: on its own it conditions nothing, and manufacturing a text_input for it would route a plain separation request carrying a language hint through the text path. The keys are left in task.options rather than erased, because vevo2's loader advertises target_text as a request option and a family reading it there keeps working. Nine tests, all confirmed failing on behaviour against a stub that returned false before the implementation was written. Verified end to end afterwards: vevo2-q8_0 with task:s2s and params[text] returns a 5.12 s 24 kHz output whose closed-loop citrinet transcription is exact, and htdemucs separation with no text param still returns its four stems, with and without params[stem]. audio-cpp-vevo2-speech-to-speech ships that route. Every audio.cpp task kind with a loadable family now has a gallery entry; spk remains the only gap and has no family upstream at all. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-5 [Claude Code] * docs(audio-cpp): document params[text] and the pinned transform tasks The text channel and the two task pins are both invisible from the endpoint contract alone: nothing in the AudioTransform form tells a reader that a speech-to-speech model needs the line it is resynthesising, and nothing says that asking for singing voice conversion without task:svc silently gets plain voice conversion instead. Both are the kind of thing a user only discovers from a refusal or, worse, from output that looks right and is not. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-5 [Claude Code] * fix(utils): annotate the two G304 sites this branch introduced gosec flags os.Open on a variable path, and both new call sites in ffmpeg.go are its alerts on this PR. Neither is reachable by an outside caller: isPCM16Wav opens the exact path it is about to hand ffmpeg as input, which in the upload path is a server-created temp file named from path.Base of the client name so no traversal survives, and wavAudioBytes opens AudioResample's own dst, a name this package derives from src and has just had ffmpeg write. Annotated in the repo's existing style rather than restructured, with the reason spelled out, because a bare suppression is worth nothing to the next reader. The three other G304 sites in this file, in passthroughWAV and isTargetWav, predate the branch and are left untouched. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-5 [Claude Code] * fix(audio-cpp): build arm64 with gcc-14 for the armv9.2 SME variants The arm64 CPU image failed to build: cc1: error: invalid feature modifier 'sme' in '-march=armv9.2-a+dotprod+fp16+sve+i8mm+sve2+sme' ggml's CPU_ALL_VARIANTS table includes armv9.2 variants compiled with +sme, and Ubuntu Noble's default gcc-13 rejects that feature modifier. Every entry in the table has to compile even though a host only ever dlopens the one its own CPU supports, so a single unbuildable variant fails the whole image. gcc-14 accepts it, which is exactly the fix llama-cpp already carries in .docker/llama-cpp-compile.sh; this is the same problem reached by a different Dockerfile. Applied to every arm64 BUILD_TYPE rather than to the CPU one alone, and that differs from llama-cpp on purpose. llama-cpp needs it only for its pure-CPU image because its GPU builds run llama-cpp-fallback, which builds no variant table. This backend's Makefile turns ENGINE_ENABLE_CPU_ALL_VARIANTS on for every non-Darwin build, GPU included, so an arm64 GPU image would hit the identical error. The matrix has no arm64 GPU entry today, which is precisely why gating on an empty BUILD_TYPE would leave the trap armed for whoever adds the first one. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-5 [Claude Code] --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-07-30 12:11:56 +02:00
Richard Palethorpeandlocalai-org-maint-bot	9058a2bb46	feat: Add 3d generation UI/API and trellis2cpp backend (#10979) ... * feat(3d): add Generate3D RPC, FLAG_3D capability, and /v1/3d/generations endpoint Adds the plumbing for image-conditioned 3D asset generation (binary glTF / GLB output), modeled on the video generation path: - backend.proto: Generate3D RPC + Generate3DRequest (staged image src, glb dst, seed/step/cfg_scale/texture_steps, quality and background enums, params map for backend-specific extras) - pkg/grpc: thread Generate3D through client, server, embed, base and the backend interfaces; connection-evicting and distributed-node wrappers (in-flight tracking + file staging) included - core/config: FLAG_3D usecase (guessed only for the trellis2cpp backend), '3d' canonical usecase string mapped to the Generate3D method, and a '3d' output modality - REST: POST /v1/3d/generations (+ unversioned alias) returning OpenAIResponse with a /generated-3d URL or b64_json; conditioning image accepted as URL, base64, or data URI; quality/background validated at the edge; .glb served as model/gltf-binary - auth: '3d' route feature (default ON); /api/instructions entry Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(trellis2cpp): add the trellis2.cpp image-to-3D backend Wraps localai-org/trellis2cpp (C++/GGML port of Microsoft TRELLIS.2, pbr-textures branch) as a Go+purego backend, following the stablediffusion-ggml pattern: - backend/go/trellis2cpp: purego bindings to the flat C ABI (v9, asserted at startup), eager pipeline load with model-set validation (refuses non-trellis GGUFs; degrades coarse/geometry-only/textured exactly like the upstream demo), Generate3D via t2_generate + t2_bake_glb writing a binary glTF to dst. Weight-free unit tests cover resolution/validation/param mapping — CI never downloads the multi-GB GGUF set or runs inference. - CPU SIMD variants build into per-variant directories (the shared libggml sonames collide across variants, unlike sd-ggml's flat renamed-.so scheme); run.sh picks one via /proc/cpuinfo. - CI wiring: backend-matrix entries (cpu, cuda12/13, vulkan amd64+arm64, l4t, l4t-cuda13, darwin metal), index.yaml meta + latest/master image entries, bump_deps tracking of the pbr-textures branch, changed-backends.js mapping, top-level Makefile targets. - Importer: auto-detects trellis GGUF repos/URIs (registered before llama-cpp so the .gguf match isn't stolen) and expands any trellis URI to the full 10-file component set spanning the three LocalAI-io HF repos. - Gallery: trellis2-4b (full PBR + 1024 cascade) and trellis2-4b-geometry (512 untextured) with verified sha256s. Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(ui): 3D generation page with native GLB viewer and IndexedDB history Adds a Studio tab + /app/3d page for the new image-to-3D endpoint: - GlbViewer ports the trellis2cpp demo's dependency-free WebGL2 renderer (quaternion trackball, metallic-roughness PBR, ACES, hidden-line wireframe with a bounded index budget) and pairs it with a minimal GLB parser for the two forms t2_bake_glb emits — dense vertex-PBR (linear COLOR_0 + _METALLIC_ROUGHNESS, uploaded as normalized integers) and the opt-in UV-atlas textured form. Parsing happens before any GL so stats and errors render without WebGL2. - use3DHistory stores past generations (params, input thumbnail, and the GLB blob itself) in IndexedDB with keep-newest-20 eviction — GLBs are multi-MB binaries localStorage can't hold — and the page offers a download button for the active GLB. - Wiring: CAP_3D capability constant (FLAG_3D — the exact string /api/models/capabilities serves), threeDApi, router entries, Studio tab, vite dev proxy, en locale keys. - e2e: render-smoke entry plus a focused spec that feeds a real one-triangle vertex-PBR GLB through the parser/viewer and exercises IndexedDB persistence, selection, deletion, and API errors. Assisted-by: Claude:claude-fable-5 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(3d): address API correctness and UX issues Keep 3D generation on the LocalAI-specific /3d/generations route and ensure authentication and permissions cover it. Propagate distributed transfer failures, publish a portable ARM64 backend image, honor importer overrides, and align discovery, upload validation, and touch controls. Assisted-by: Codex:gpt-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(3d): add previewable print remeshing Add a single-detail CGAL Alpha Wrap workflow for existing Trellis GLBs, including PBR reprojection, API documentation, tracing, and an in-browser preview before download. Allow the remesh route to enforce its 512 MiB upload cap independently of the smaller global default so generated high-resolution meshes can be processed. Assisted-by: Codex:gpt-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> * build(trellis2cpp): centralize remesh dependency pins Assisted-by: Codex:GPT-5 [apply_patch] [exec_command] Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(kokoros): implement Generate3D stub for new proto RPC The Generate3D RPC added to backend.proto for the trellis2cpp backend made tonic's generated Backend trait require generate3_d, breaking the kokoros-grpc build. Return unimplemented like the other unsupported modalities. Assisted-by: Claude Code:claude-fable-5 Signed-off-by: Richard Palethorpe <io@richiejp.com> --------- Signed-off-by: Richard Palethorpe <io@richiejp.com> Co-authored-by: localai-org-maint-bot <bot-opensource@localaisrl.com>	2026-07-29 16:15:04 +02:00
Tai An	a7fa678d83	fix(tts): forward the OpenAI speed field to the backend (#11097) (#11120) ... * fix(tts): forward the OpenAI speed field to the backend (#11097) /v1/audio/speech accepted the documented OpenAI `speed` field and then dropped it: schema.TTSRequest had no Speed member, so the value never reached proto.TTSRequest and the request returned 200 with an unchanged playback rate. Accept speed and normalise it into the existing per-request params map, which core/backend forwards verbatim to the backend. An explicit params["speed"] still wins, and a value outside the documented 0.25-4.0 range is now rejected with 400 instead of being silently ignored. Signed-off-by: Anai-Guo <antai12232931@outlook.com> * fix(tts): distinguish explicit speed=0 from an omitted field Make TTSRequest.Speed a *float32 so an explicit `"speed": 0` (invalid, below the documented 0.25 minimum) is rejected with 400 instead of being treated as unset and silently defaulted. An omitted field stays nil and leaves the backend default untouched. Add a request-boundary regression that distinguishes an omitted speed from an explicit zero, addressing review feedback. Signed-off-by: Anai-Guo <antai12232931@outlook.com> * docs: drop the speed field from the TTS docs Per review: no backend consumes params.speed today, so documenting it would be misleading. The API-level plumbing and validation stay. Signed-off-by: Anai-Guo <antai12232931@outlook.com> --------- Signed-off-by: Anai-Guo <antai12232931@outlook.com>	2026-07-27 19:03:32 +02:00
mudler's LocalAI [bot]andEttore Di Giacinto	83a0f16a21	feat(gallery): let one gallery entry offer several builds of the same model (#10943) ... * feat(system): expose raw detected capability for model meta resolution Model meta gallery entries express hardware fallback through candidate ordering rather than a capability map, so they need the undecorated detected capability string without Capability's default/cpu fallback chain. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * refactor(system): drop duplicate capability accessor, cover DetectedCapability ReportedCapability was added with a body identical to the existing DetectedCapability. Keep one accessor and move the specs onto it, since DetectedCapability had no direct coverage of its no-fallback behavior. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(vram): parse IEC binary size suffixes (KiB..PiB) ParseSizeString accepted only SI suffixes, so a "20GiB" floor was rejected outright. Model and VRAM sizes are conventionally quoted in IEC units, and silently reading GiB as GB would understate a floor by about 7%. Purely additive: these inputs previously returned an unknown-suffix error. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): add Candidate type for meta model entries Candidate is one option in a meta entry's ordered variant list. It names a concrete gallery entry and declares when that entry suits the host. EffectiveMinVRAM resolves the VRAM floor, letting an authored min_vram win over a nightly-inferred one. An unparseable floor errors instead of being treated as absent: swallowing a typo would turn a constrained candidate into an unconstrained one and select a too-large variant rather than fail loudly. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): add hardware-aware model variant resolver Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): allow gallery model entries to declare variant candidates A gallery entry with a non-empty candidates list is a meta entry: it names an ordered list of concrete entries and resolves to the first one the host can satisfy, instead of describing model files directly. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): resolve meta model entries to hardware-appropriate variants at install Meta gallery entries carry an ordered candidate list; at install time the first candidate the host satisfies is resolved and its payload installed under the meta's name, so the model keeps a stable name regardless of which variant backs it. The resolution is recorded in the installed gallery config so a reinstall honors a prior pin and operators can see the backing variant. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(gallery): key meta pin recall on the installed name and detach resolved entries Six review findings on the meta-entry install path. Pin recall was keyed on the gallery entry name while applyModel writes the record under the install name (req.Name when supplied), so a meta installed under a custom name with a pin lost that pin on reinstall and was silently re-resolved onto a different variant, possibly swapping its backend. Compute the install name with applyModel's own precedence before the recall. ResolveMetaModel returned a shallow struct copy, so the resolved entry's Overrides aliased the gallery entry's map and the install path's in-place mergo merge wrote the caller's request into the shared catalog. Detach Overrides, ConfigFile, AdditionalFiles, URLs and Tags. Not exploitable today only because this path re-unmarshals the gallery per call, which is a property nobody should have to rely on. Also: overlay the meta's name onto the persisted config for meta installs so the gallery file no longer records the variant's name; move the pinned-VRAM warning below the variant validation so a pin naming a nonexistent entry does not warn about VRAM before failing for an unrelated reason; and stop seeding config.URLs in the config_file branch, which duplicated every declared URL. Add seven network-free specs driving InstallModelFromGallery with a meta entry: variant payload wins over the meta's legacy url fallback, the resolution record round-trips to disk, a pin is recorded and honored on reinstall including under a custom install name, and the resolved entry does not alias the gallery's maps. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(gallery): deep-copy meta overrides and make two specs functional ResolveMetaModel detached the resolved entry's Overrides and ConfigFile with maps.Clone, which only copies the top level. Gallery overrides are nested in practice (parameters.model is near-universal) and the install path merges the caller's request with mergo.WithOverride, which recurses into nested maps and overwrites them in place, so the gallery entry's own inner maps were still reachable and still got rewritten by the last caller to install. Copy both maps all the way down instead, recursing through the container shapes a YAML decoder produces. ConfigFile is not mutated on the install path today, but it carries the same kind of nested payload and leaving it shallowly cloned would invite the bug back. Also fix two specs that passed whether or not their target fix was present: - "does not write the caller's overrides back into the gallery entry" re-read the catalog from disk, which re-unmarshals fresh structs and so cannot observe in-memory aliasing. It now asserts against the in-memory gallery entry and drives the real mergo merge. - "round-trips the resolution record to disk under the meta's name" asserted a name that is already correct in the config_file branch. It now drives the url branch via a file:// fixture, where the meta-name overlay actually applies. Both were verified red by reverting their fix. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(gallery): lint meta model entry invariants in index.yaml Adds Ginkgo specs that parse the shipped gallery/index.yaml and enforce the invariants that keep meta entries safe: a legacy url fallback equal to the final candidate's url, references only to existing non-meta entries, a min_vram floor on every candidate but the last-resort one, a capability drawn only from the vocabulary the system can report, and descending VRAM floors within a capability group. The capability check is the only compensating control for a typo there. Candidate matching is a case-sensitive exact comparison against SystemState.DetectedCapability(), so an unknown value never matches and falls through silently instead of erroring. The vocabulary therefore mirrors the raw return set of getSystemCapabilities(), which notably excludes "cpu": that is a fallback key inside Capability(capMap) on the meta backend path, never a reported capability. A CPU-only host reports "default". These pass vacuously until the pilot meta entry lands; the guard is intentionally in place before the thing it guards. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(gallery): close coverage gaps in the meta entry lint The ordering invariant grouped candidates by capability and asserted floors descend within a group. A candidate with an EMPTY capability matches every host, so it does not belong in its own group: it dominates every later candidate whose floor is at or above its own, across capability groups. Track a running minimum floor over the unconditional candidates instead, which subsumes the old same-group check for the empty capability. Every spec skipped non-meta entries, so with zero meta entries in the index all five bodies were no-ops. Aligning GalleryModel.IsMeta() with GalleryBackend.IsMeta(), whose semantics are deliberately opposite, would have made all of them pass while checking nothing. Extract each invariant into a helper over a slice of entries returning the violations it finds, and cover those helpers with synthetic fixtures so the logic stays tested at zero meta entries. The index-driven specs are now a thin application of already proven logic. Also assert the index parses non-empty, report every violation in one run rather than aborting on the first, and parse the index once for the suite. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * ci(gallery): add nightly denormalization of meta model candidates Fills the read-only backend, quantization and inferred_min_vram fields on meta gallery candidates and opens a PR, modeled on the existing checksum_checker job. Computing these needs network access, so it happens nightly rather than at install time. An authored min_vram is never modified: a human who measured a real load knows more than a pre-download estimate does. The index is rewritten via yaml.Node rather than a document round-trip. A full round-trip reflows all ~26k lines of gallery/index.yaml, which would bury the computed values and make the nightly PR unreviewable. The rewrite touches only the three derived keys, so authored styling survives and a run that computes nothing leaves the file untouched. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ci): keep the gallery denormalize diff reviewable and self-healing The nightly denormalization job edits YAML nodes instead of round-tripping structs so its PR stays small enough for a human to review, but the write path undid that: yaml.Marshal re-encoded the node tree at yaml.v3's default 4-space indent and dropped the leading document marker, reflowing roughly 6000 lines around the handful of real changes. Encode through yaml.NewEncoder at the index's authored 2-space indent and restore the header. A write that changes three fields now changes three lines. Stale inferred_min_vram values were also never cleared. Both skip paths (an authored min_vram is present, or the candidate is the last resort) returned before touching the field, so a candidate that gained a floor or became the last resort after a reorder kept an inferred value that EffectiveMinVRAM reported as a real constraint, failing the meta lint with no way for the job to self-heal. Clear the field before both skips. The workflow discarded a whole night's work on any single failure: the program exits 1 when a candidate cannot be estimated, which aborted the job before the PR step, so one unreachable candidate blocked every other refresh indefinitely. Capture the status, open the PR with what was computed, mark the PR body as partial, and fail the run afterwards so the problem still surfaces. Also preserve the index's existing file mode instead of forcing 0644, and drop the redundant //go:build ignore tag, since Go already skips dot directories and the sibling modelslist.go carries no tag. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): add nanbeige4.1-3b meta entry with hardware-resolved variants Adds the first real meta entry to the gallery index. It resolves to the Q8_0 build on hosts with at least 6GiB of VRAM and to the Q4_K_M build everywhere else, installing either payload under the stable name nanbeige4.1-3b. The entry carries a url equal to its final candidate's url. LocalAI releases that predate candidates support parse the index non-strictly and drop the key silently, so without that url they would list the entry and install nothing. A regression spec parses the index the way those releases do and asserts every meta entry stays installable for them. Also teaches core/schema/gallery-model.schema.json about candidates. The schema sets additionalProperties: false at the top level, so an author following CONTRIBUTING.md and adding the yaml-language-server comment would otherwise get a validation error on this entry. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): make candidate entries complete, installable entries Reworks hardware-resolved gallery variants after a design pivot. There is no longer a separate "meta" entry kind. A gallery entry is a normal, complete entry that may additionally carry candidates:, a list of hardware-gated upgrades over itself, and the entry is itself the last-resort candidate. The previous design relied on a bare url: as the fallback for LocalAI releases that predate candidates support. That fallback is empty in practice: none of the 80 gallery/*.yaml files carry a top-level files:, and 1216 of 1281 index entries carry their payload in the index entry itself, so a url alone yields a config template with nothing to download. Since every released LocalAI reads gallery/index.yaml live from master, merging a payload-less entry would have shown every existing user a model that installs to a broken state. Making the entry its own base candidate removes the problem at the root: old clients drop the candidates key and install the entry exactly as they do today. Resolution order is now explicit pin, then capability plus VRAM over the declared upgrades, then the entry itself. The entry ALWAYS installs: when its own min_vram or capability is unmet the installer warns and installs it anyway, because there is nothing below it and refusing would make the gallery behave worse the newer the client is. A pin naming the entry's own name is valid and is how an operator declines an upgrade. IsMeta() becomes HasCandidates(), ResolveMetaModel becomes ResolveVariant, and the persisted meta_name record key becomes entry_name. GalleryBackend.IsMeta() is a separate concept and is untouched. The lint drops the three rules the pivot makes wrong (url equality with the final candidate, no inline payload, unconstrained final candidate) and gains one: the entry's own floor must sit strictly below every candidate's, since a base that outranks a candidate makes that candidate unreachable. The pilot entry is now the existing nanbeige4.1-3b-q4, which gains a 2GiB floor of its own and a single 6GiB upgrade to nanbeige4.1-3b-q8, replacing the separate nanbeige4.1-3b entry added in d0d441bb4. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): select model variants by hardware fit, not authored order Gallery entries could already carry a list of alternatives, but selection was an authored, ordered, first-match policy: every candidate declared a `capability` string and the VRAM floors had to descend in a hand-tuned order. That pushed hardware knowledge onto whoever edits the gallery and made ordering load-bearing, so a reordered list silently changed what users installed. None of it was necessary. SystemState.IsBackendCompatible already derives hardware support from a backend name alone: it knows MLX and metal are Darwin-only, CUDA is NVIDIA-only, ROCm AMD-only, SYCL Intel-only. Selection can read that instead of asking authors to restate it. Authoring is now just a list of names: - name: qwen3.6-27b min_memory: 4GiB variants: - model: qwen3.6-27b-mlx-8bit - model: qwen3.6-27b-gguf-q8 min_memory: 28GiB and all the intelligence moved into the selector. Given a host it drops the variants whose backend cannot run here, drops those whose known memory requirement exceeds what the host has, and takes the LARGEST of what is left, because a bigger footprint is a higher quality quantization of the same model. A variant of unknown size is kept, since nothing proves it does not fit, but it ranks last so a proven fit always beats a guess. An explicit pin still wins outright, and if nothing survives the entry installs its own payload: the base always installs, this never refuses. Available memory is VRAM when a GPU was detected and system RAM otherwise, read through xsysinfo so a cgroup limit is honored and a container gets its own limit rather than the node's RAM. Capability disappears entirely, from the types, the schema and the lint. VRAM and RAM collapse into one `min_memory`, because a model's footprint is roughly the same wherever it lives and one figure is compared against whichever applies. The lint rules about ordering, the capability vocabulary and floor relationships are deleted with the hazards they described; what remains is that every variant names an entry that exists and does not itself declare variants, plus that any memory figure actually parses. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * gallery: size model variants with a live probe, drop the nightly denormalizer Selection needs each variant's size to decide whether it fits and to rank largest-first. That figure was written into the index by a nightly job, which made the gallery carry a derived value that could drift from the entry it was derived from. Derive it at install time instead. pkg/vram already sizes a model without downloading it, and the gallery UI already uses it: a remote GGUF header range-fetch, then an HTTP HEAD for the content length, then any declared size:. It caches its results, so reuse it rather than writing a second probing path. A probe failure must never fail an install, so an unprobeable variant is treated as unknown: it survives the memory filter, because nothing proves it does not fit, and it ranks last, so a known-good fit always beats a guess. If every probe fails, selection still terminates on the base entry. The probe is injected through ResolveEnv rather than called directly, for the same reason the backend compatibility check is: specs pin an exact size, or an exact failure, without reaching the network. With that in place three things are dead weight and go: - The nightly job and the fields it populated. Variant.Backend was redundant because the backend is resolved live from the referenced entry during selection, and Quantization was display-only that nothing read. - min_memory on the base entry. The base always installs and its floor could only warn, so it could not change any outcome. - The lint rules and schema entries for both. min_memory on individual variants stays, as the override for when the probed size is wrong. An authored figure now suppresses the probe entirely rather than merely outranking it, so it costs no round trip. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): expose model variants for selection over API, CLI and MCP A gallery entry may carry `variants:`, alternative builds of the same model. Selection already worked at install time, but nothing could see what an entry offered or ask for a specific build, so the feature was undrivable. Listing: `GET /api/models` now reports `variants` and `auto_variant` for the entries that declare variants. Each variant carries its resolved backend, its measured size and whether it fits this host. `auto_variant` is what installing without a choice would pick right now. The new gallery.DescribeVariants runs the same variantOptions + SelectVariant pass the installer runs, so the reported default cannot drift from what installing actually does, and HostResolveEnv is extracted so both derive the host and share pkg/vram's probe cache from one place. Performance: an entry that declares no variants returns early without touching the probe, so the ~1280 ordinary entries cost exactly what they cost before. Selection: `variant` is accepted on POST /models/apply, as a query param on POST /api/models/install/:id, on the gallery apply file/string request, as `local-ai models install --variant`, and as a parameter on the install_model MCP tool (both the httpapi and inproc clients). Empty means auto-select. An unknown variant name now fails the install naming what was requested. This closes a real hole: an entry declaring no variants short-circuits before selection runs, so a requested variant was previously dropped silently and the install reported success. startup.InstallModels ends in a variadic model list, so install options could not be appended to it; InstallModelsWithOptions is added alongside and InstallModels delegates to it. No caller signature changed. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): drop the redundant variant min_memory field Variant.MinMemory was an authored override for when the live probe misreads a variant's footprint. It duplicated an existing field: probeEntryMemory already passes the entry's declared size: into EstimateModelMultiContext, whose cascade prefers that declared size over its own guesswork. Correcting size: on the referenced entry fixes the figure for every consumer rather than only for variant selection, so min_memory shadowed the right answer. A variant is now nothing but a name. Its effective size is exactly the probe result, and an unknown stays unknown: it survives the filter and ranks last. EffectiveMemory loses its error return along with the field. The authored string was the only thing that could fail to parse, so the error had no remaining source and was propagating dead nil-checks through SelectVariant, DescribeVariants and the pin warning. Selection behaviour is unchanged. The specs covering probe-derived sizing, ranking, filtering, the unknown-size path, pin recall, entry/variant metadata split and deep-copy isolation all survive; the three install specs that needed a definite size now declare it through the referenced entry's own size:, which exercises the documented escape hatch directly. gallery/index.yaml is untouched: no entry ever carried the key. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(gallery): rank the entry's own build against its variants Variant selection pulled the declaring entry's own payload, the base, out of the candidate set and consulted it only once every declared variant had been rejected. Two real failures followed. A variant whose size the probe cannot determine deliberately survives the memory filter, because nothing proves it does not fit. As the only survivor it then won outright on any host, however small: a 2GiB machine installed an unmeasured variant in preference to the 4GiB build the entry itself ships, with no warning. 241 of the 1280 current index entries carry no files and no size, which is exactly that shape. "Largest wins" also broke whenever the base was the largest. An author writing a Q8 entry that offers a Q4 downgrade for small hosts, a natural shape that nothing in the lint, schema or docs discourages, had the Q4 installed on every large host instead. Make the base an ordinary participant. It is still exempt from both filters, so selection always terminates on something installable, but it is now ranked against the variants: a proven fit first and largest, then the base, then any variant whose size nothing could measure. Both failures disappear together. The base is probed for its size accordingly, which it was not before, because an unsized base would lose every contest to an unmeasurable variant. FellBackToBase is kept but narrowed to "no declared variant survived", rather than "the base was chosen", since the base now also wins on merit and that is not worth warning about. A recalled variant pin also became a permanent install failure. A pin the caller supplies on this request must stay fatal, but one recalled from ._gallery_<name>.yaml can be invalidated by any later gallery edit, and failing on it turned one rename into a model that could never be reinstalled or upgraded again short of deleting a dotfile the user has never heard of. A stale recalled pin is now dropped with a warning naming it, and selection runs as if it had never been recorded. Also drop the last textual reference to two abandoned designs from the DetectedCapability comment, correct the documented variants JSON example, which showed a memory_bytes of 0 that omitempty makes impossible, and remove an em dash from the install skill. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(gallery): budget variant memory from RAM when a GPU reports no VRAM Variant selection read its memory budget from VRAM whenever a GPU capability was detected, and from system RAM only when none was. Apple Silicon satisfies the first branch and fails the premise: arm64 macs report the metal capability unconditionally, without probing anything, while TotalAvailableVRAM has no discrete VRAM pool to find and returns zero. The budget therefore came out as zero on every Mac. Zero drops every variant carrying a known size, so the base build was installed on all of them however much memory the machine had. The feature was inert on the platform, and silently: falling back to the base is a legitimate outcome, so nothing looked wrong. Take VRAM only when it is actually a number, and fall back to RAM otherwise. On a unified-memory host RAM is not an approximation of the budget, it is the budget, since the GPU shares it. A discrete GPU whose VRAM could not be read also lands on RAM, which overstates what the card holds but understates nothing the host has; the previous zero understated both. An unreadable RAM figure still yields zero and still installs the base, so a genuinely unknown host is not talked into a larger download. This is what turned tests-apple red: "installs a fitting variant's payload under the entry's own name" asserts on selection, and the runner resolved to the base because its budget was zero. The added specs pin the branch directly rather than relying on a macOS runner to notice again. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): add a model variant picker to the models gallery PR #10943 shipped the server side: a gallery entry may declare `variants:`, `GET /api/models` attaches `variants` and `auto_variant` to declaring entries, and `POST /api/models/install/:id` accepts a `variant` query parameter. Nothing in the UI consumed any of it, so the feature was not reachable from the browser. This wires it up. modelsApi.install takes an optional second argument and appends an encoded `?variant=` only when one is given, so every existing call site keeps sending exactly the request it sent before. On the models table, an entry that declares variants gets a split button. The primary Install still installs the auto-selected build, because auto is the default and the point of the feature; the chevron opens a menu for a deliberate override. It follows the Backends.jsx precedent: one shared Popover re-anchored per row, rendering .action-menu items, which brings Escape, outside-click and focus return along with it. An entry that declares no variants renders exactly as it did before. A variant that does not fit is dimmed but stays selectable, since the server honors an explicit choice with a warning rather than refusing it. memory_bytes is omitempty on the wire, so an absent key means the size is unknown and never zero. A single helper guards both the menu and the detail row, because formatBytes would otherwise render a falsy value as "0 B", which reads as "needs nothing". The expanded detail row gains a Variants section listing each build's backend, size, whether it fits, which is the entry's own build, and which one auto-selection would pick, built from the existing DetailRow helper and .badge classes. Eight Playwright specs cover the picker, including that plain Install sends no variant parameter and that choosing one sends it. One pre-existing assertion was scoped with .first(): the Variants section legitimately adds more llama-cpp badges to the detail row, which tripped strict mode. UI line coverage 49.42% -> 49.36% against a 40.0 baseline and 0.8pp tolerance; branch coverage rose 72.04% -> 72.66%. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(gallery): describe model variants from a companion endpoint Variant description probes each referenced entry's weight files over the network: an HTTP HEAD plus a ranged GET, serial, five seconds per probe with no aggregate deadline. Running it inline in GET /api/models made one listing cost (entries x variants) round trips. The Manage page fetches with items=9999, so at 200 declaring entries that is ~1000 serial probes, minutes of a blocked handler and gigabytes of range traffic for a single page load. Only one entry declares variants today, but the feature exists so that many will. Follow the precedent already set for VRAM estimates. The listing now reports only has_variants, a length check on loaded metadata that touches nothing, and GET /api/models/variants/:id returns the description for one entry, mirroring estimate/:id in route shape, auth and error handling. DescribeVariants itself is unchanged; only its caller moved. The picker fetches lazily at the two points where a user asks to see variants, opening the split-button menu and expanding the detail row, and caches per entry for the page session. An entry declaring no variants issues no request at all. A spec counts real HTTP hits on the weight files, so it goes red if description becomes reachable from the listing path again through any caller. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): filter the model gallery to entries that declare variants The gallery is heading towards showing parent entries and hiding the individual builds they reference, so a user sees one row per model rather than six quantizations of it. Adoption is a single entry today, so defaulting to that would leave a one-row gallery. This ships the migration-phase inverse instead: the default is untouched, and a toggle narrows the list to only the entries that declare variants. It previews the end state and changes nothing until someone asks for it. The filter is server-side, next to term/tag/backend/capability and above the pagination arithmetic. The listing paginates at 9 items, so narrowing on the client would leave totalPages and availableModels describing the unfiltered set and hand the user empty pages. It selects on HasVariants(), which reads already-loaded metadata, so it issues no variant probes. The parameter is named has_variants after the listing field it selects on, and is compared against "true" like the other boolean query params (all_users, save_checkpoint), so has_variants=false reads as absent. With it omitted the response is byte-for-byte what it was before. The control is the shared Toggle component, matching the fitsFilter toggle already on this page: same wrapper class, same icon and label shape, same localStorage persistence. Unlike fitsFilter it resets to page 1 on change, which a server-side filter has to do. Stacking the toggle with a tag or backend filter easily yields nothing while one entry declares variants, so the empty state now names the variants filter as the cause rather than leaving a user to conclude the gallery is broken. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): render gallery model descriptions as Markdown Gallery descriptions are Markdown, but the React UI dumped them raw, so a model whose description opens with an ATX heading showed a literal "# Qwen3.6-27B [](https://chat.qwen.ai)" in the list. Full-description areas now render through renderMarkdown (marked + DOMPurify), matching how Backends.jsx and the Manage detail panels already handle the same content: - Models.jsx expanded detail row - VoiceLibrary.jsx voice detail header The truncated one-line previews must not render block Markdown: a leading "#" would become an <h1> and wreck the row height and rhythm. They get a new stripMarkdown() helper instead, which reduces Markdown to a single line of readable plain text. It is used for the cell text and for the title tooltip, since a tooltip full of "[](url)" is no better than a cell full of it: - Models.jsx gallery table description cell - Manage.jsx model and backend resource-row descriptions stripMarkdown walks marked's lexer output rather than running regexes over the source, so what it strips is by construction what renderMarkdown would have rendered, and it needs no new dependency. Output lands in JSX text nodes, so React escapes it; no new dangerouslySetInnerHTML beyond the two full-description sites, both of which run DOMPurify. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ui): strip Markdown from the backends table description cell Commit b35d630cf fixed this for gallery models but left the Backends admin page with the same asymmetry: its detail panel renders the description through renderMarkdown, while the collapsed table row dumped the raw gallery string into both the cell body and the title tooltip. That is user-visible. 40 of the 949 entries in backend/index.yaml carry Markdown - insightface uses inline code backticks, others use lists and links - and backend descriptions also contain embedded newlines, so the one-line cell showed literal syntax. The cell now runs stripMarkdown over the description once and uses the result for the text and the title, matching Models.jsx and the ResourceRowDesc component in Manage.jsx. The '-' placeholder is preserved, and now also fires when a description reduces to nothing after stripping. The detail panel is untouched and no new dangerouslySetInnerHTML is introduced: stripMarkdown output lands in a JSX text node, so React escapes it. Three Playwright specs cover it: a description with a heading, inline code and a link renders as clean text with no literal syntax and no block element in the cell, the title tooltip carries the same stripped text, and a backend without a description still shows the placeholder. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * ui(models): polish the variant detail view and scope rendered Markdown The gallery detail pane rendered every field through the same two-column label/value row, including the description. Multi-paragraph prose in a value cell ran eight rows tall at the top of the pane on a ~1200px measure, breaking the grid's rhythm exactly where the eye enters. Move it into its own full-width block above the table, capped at a 68ch measure, keeping the label. Rendered Markdown had no scoped typography anywhere in the app, so a description opening with `#` inherited the browser default 2em inside a 13px surface while a `##` further down was indistinguishable from body text. Add a reusable .markdown-body block mapping h1-h6, paragraphs, lists, links, code, blockquotes, images and tables onto the existing type scale, and apply it to every renderMarkdown() consumer: the models detail, the backends detail, both Manage details and the voice library detail. Rebalance the variants list so the name leads. Backend and size drop from badge/secondary weight to muted metadata; the FITS badge goes entirely, since it was true of nearly every row and so said nothing, while the variant that does not fit keeps a warning badge and a dimmed name. AUTO-SELECTED stays marked because it answers what a plain Install produces. Rows share the parent's grid tracks via subgrid so name, backend, size and status line up down the list instead of raggedly following name length. Finally, make each variant row actionable. It looked like a list of choices but was inert text, with per-variant install hidden behind the split-button chevron elsewhere; each row is now a button onto the existing handleInstall(modelId, variant) path, with hover, keyboard focus and a disabled state while an install is in flight. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): collapse the listing to one row per model The listing supported has_variants=true, which narrowed to entries that DECLARE variants. With adoption at three entries that showed three rows, which is useless; it was always a placeholder. Replace it with the view that is actually useful: the deduplicated gallery. Show every entry installable in its own right and nothing twice, which means the parents plus every entry nobody references, and hide only the builds another entry already offers as a variant, since those are reachable through their parent. The parameter is renamed to collapse_variants accordingly: the filter is no longer a predicate on a row's own metadata but a view over the whole gallery. Default stays off, so the response with the parameter absent is unchanged. VariantReferencedIDs never reports an entry that declares variants of its own, so parents are always visible. That guarantees every hidden entry has a visible entry offering it, and no chain can strand a row. Variant resolution already refuses to install such a reference, but the listing has to stay coherent in the presence of a gallery that has one rather than silently swallowing entries. Self-references and dangling references hide nothing. The referenced set is computed over the whole gallery rather than over what the other filters left, so an entry is hidden because a parent offers it and never because of what the user searched for. The pass is over metadata already in memory: it resolves nothing over the network and triggers no variant description or size probe, so the listing's zero-probe contract still holds. The UI toggle keeps its behaviour (persistence, page reset, clear filters) and becomes "One row per model", which says what the user gets. Its localStorage key moves too, since the stored value meant a different filter. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): show the collapsed model listing by default The gallery listing is what a user reaches for to answer "what can I install". Answering that with several rows for the same model, one per build, makes the reader do the deduplication the collapsed view already does, so the collapsed view is the one to land on. The UI now asks for collapse_variants=true unless the toggle says otherwise. The server default is deliberately untouched: a request with the parameter absent still returns the full listing, because other API clients depend on that response and collapsing it under them would be a breaking change. Opting out omits the parameter rather than sending false, so it asks for exactly the listing everyone else gets. The stored preference changes vocabulary from '1'/'0' to 'on'/'off'. The previous build wrote it from an effect that runs on mount, so a stored '0' recorded that the page had been opened rather than that anyone chose the expanded view, and honouring it would pin every earlier visitor to a default they never picked. Only the new vocabulary counts as a choice; a legacy '1' meant the collapsed view and is what the new default gives anyway, so no earlier deliberate choice is lost. Collapsing being the default also changes what the empty state may say about it. An opted-into filter can be named as the cause of an empty result; a default cannot, so the filters keep the top line and the collapsed view drops to a hint below it, shown only once filters are narrowing the set. For the same reason "Clear filters" now restores the collapsed default instead of switching it off, and the toggle alone no longer counts as a filter worth offering to clear. The label stays "One row per model": it describes the view the user is looking at rather than an action, so it reads the same whether it is opted into or out of. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * gallery: group alternative builds of the same weights under variants Sweep the gallery for entries that are alternative builds of the same weights (different quantization, precision, or runtime format) and declare them as variants of a single parent row, so the listing offers one row per model instead of one row per quantization and the installer picks the largest build that this host can actually run. 41 families over 95 entries, turning 54 entries into variants. The parent is the bare-named entry wherever one exists, so nothing changes about what any existing entry installs. Ranking already selects the largest fitting build regardless of which entry is nominally the parent, so the parent only decides the pathological case where nothing fits. For the ten families that have no bare-named entry, the smallest build is the parent, since that is the one that has to install when nothing fits. Grouping was verified against the actual model filenames rather than the entry names alone. Different parameter sizes, languages, finetunes, and products that merely share a name prefix are left as separate rows: the qwen3.6 APEX and pi-tune finetunes, the DFlash and MTP speculative-decoding pairings, English-only versus multilingual Whisper, the QAT versus non-QAT Gemma 4 weights, and the abliterated FLUX build are all distinct models. Six parents define YAML anchors that other entries pull in with a merge key, which would have handed their variants to every merging child. For the two depth-anything anchors that would have made fourteen unrelated entries advertise the base model's builds as their own. All 26 merging children therefore carry an explicit empty variants list, which overrides the merged key and is equivalent to the key being absent. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): rank model variants by host backend preference Variant auto-selection filtered candidates by whether their backend can run on the host, then ranked the survivors by size alone. The backend never influenced the choice beyond that gate, so a Mac offered both an MLX build and a llama.cpp build kept neither filtered and installed whichever was larger, leaving the native accelerated runtime unused. The same held for CUDA against CPU on NVIDIA and ROCm against Vulkan on AMD. Rank by the host's backend preference between the fit tier and size: fit stays a filter, preference decides among the builds the host can equally hold, and size still separates builds on equally preferred runtimes. The preference data stays in one declarative table in pkg/system, now read by a prefix lookup instead of a switch, so adding a capability or reordering one host's runtimes is a one-line edit and the gallery's ranking code carries no per-backend branching. MLX joins the metal rule ahead of metal itself, which is inert for the existing alias-resolution consumer because no alias group holds a candidate named for mlx. An unrecognised backend, an unrecognised capability and an absent preference list all collapse to the previous size-only ordering rather than erroring or dropping candidates. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(gallery): rank variants by engine name, not backend build tag Variant auto-selection ranked candidates with SystemState.BackendPreferenceTokens, but that function and the variant ranker speak different vocabularies. BackendPreferenceTokens returns BUILD TAGS ("cuda", "rocm", "sycl", "vulkan", "metal", "cpu"). It exists to match installed backend build directory names like "llama-cpp-cuda-12" during alias resolution in ListSystemBackends. Variant ranking instead matches a gallery entry's `backend:` value, which is an ENGINE NAME: "llama-cpp", "vllm", "vllm-omni", "sglang", "mlx" and the rest. No engine name in gallery/index.yaml contains "cuda", "rocm", "sycl" or "vulkan". preferenceRank matches by substring, so on an NVIDIA host the tokens [cuda, vulkan, cpu] matched neither "llama-cpp" nor "vllm", every candidate scored identically and size alone decided. The NVIDIA, AMD, Intel, darwin-x86 and vulkan rules were all inert. Only metal appeared to work, and only because the token "mlx" happens to equal an engine name. The mismatch does not error, it silently deletes the feature. Separate the two vocabularies. backendBuildTagPreferenceRules keeps the build tags and its original output for every capability, including metal, whose "mlx" token is removed again; its alias-resolution consumer is byte-identical to before. engineNamePreferenceRules is new, holds engine names, and is read by the new EnginePreferenceTokens, which HostResolveEnv wires into the renamed ResolveEnv.EnginePreference. Both tables sit adjacent under one block comment naming each vocabulary and each consumer, and share one lookup helper so their semantics cannot drift. On NVIDIA the order is vLLM, then SGLang, then llama-cpp: vLLM is the throughput engine and a model published with a vLLM build is published that way because that build is the one worth running. AMD and Intel get the same order, since rocm and intel builds of both serving engines ship. Metal prefers mlx over llama-cpp. Vulkan prefers llama-cpp, the only LLM engine with a Vulkan build. darwin-x86 and unknown capabilities are deliberately absent rather than guessed at, degrading to the size-only ordering that predates preference. preferenceRank stays generic and names no engine and no capability, so adding a runtime remains a one-line table edit. Specs pin the NVIDIA and metal rules through the live table and the real HostResolveEnv wiring, so emptying the engine table or wiring the build tag source back in both go red. A regression table asserts BackendPreferenceTokens' original output per capability, and mirrored locks assert neither table carries the other's vocabulary. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: record that variant selection ranks by engine before size A gallery entry can now declare variants, and selection ranks the builds a host can run by engine preference before size. Nothing told a contributor adding a backend that engineNamePreferenceRules exists, so a new engine would silently rank below every known one and lose to whatever build happened to be larger on hosts where it should have won. Document the step where a backend is added, warn against the sibling backendBuildTagPreferenceRules table (build tags, not engine names: the wrong table matches nothing, scores every candidate equally and disables the preference without erroring), and index it from AGENTS.md. Fix the authoring and user docs, which still claimed the largest surviving build wins. An author grouping builds under one entry has to be able to predict what a user gets, and size alone no longer decides it. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(cli,mcp): describe variant auto-selection as preference before size The CLI flag help and the install_model tool schema both still said auto-selection takes the largest build that runs. Ranking now puts engine preference ahead of size, so on NVIDIA a vLLM build wins over a larger llama.cpp one. An assistant reading the old schema would tell users the wrong thing. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(gallery): prefer llama.cpp over GPU serving engines on hosts with no GPU engineNamePreferenceRules had no row for the "default" capability, which getSystemCapabilities() returns both when no GPU is detected and when a GPU is present but under the 4 GiB VRAM floor. A missing row yields an empty preference list, which preferenceRank reads as "score everything equally", collapsing variant selection to size alone. That would be harmless if the hardware filter dropped GPU serving engines on such a host, but it does not. IsBackendCompatible derives support from the engine NAME, and "vllm" and "sglang" contain none of the darwin, cuda, rocm or sycl tokens it keys on, so they fall through to its closing "return true". A vLLM variant therefore survives on a CPU-only box and wins whenever its build is the larger of the two on offer: the machine installs vLLM in preference to llama.cpp. darwin-x86 had the identical hole. It was documented as a deliberate omission because nothing accelerates on an Intel Mac, which is true about acceleration and wrong about consequence: with every engine tied, download size decides. Add rows for both putting llama-cpp first. The GPU engines are enumerated behind it rather than left unmatched: an unmatched engine already ranks below every listed one, so llama.cpp would win either way, but unmatched engines also tie with each other and let size decide among them. Naming them fixes that order. MLX is left off the darwin-x86 row on purpose so it ranks last, since IsBackendCompatible admits darwin-tokened engines on that capability even though MLX needs Apple silicon. Preference orders survivors and never filters, so a model published only as a vLLM build is still installed on a host with no GPU; there is a spec for it. Surveyed every other value getSystemCapabilities() can return. nvidia, amd, intel and vulkan have rows; the l4t and cuda-refined values reach the nvidia row by prefix; "apple" and "" cannot reach the vendor fallthrough because the darwin and no-GPU branches return earlier. default and darwin-x86 were the only live holes. BackendPreferenceTokens and its build-tag table are untouched, and preferenceRank stays generic, naming no engine and no capability. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * gallery: prefer speculative-decoding builds when they fit Rank serving features between engine preference and size, so a host that can hold a DFlash or MTP build of a model's weights installs it instead of the plain build. Both answer faster for the same output, so whenever one survives the filters there is no reason to take the plain build. Precedence is now fit, then engine, then serving feature, then size. Engine outranks the feature deliberately: a serving feature makes the right engine faster, it does not make a wrong engine right, so a plain vLLM build still beats a DFlash llama.cpp build on NVIDIA. Fit outranks both, and a drafter pairing is strictly larger than the plain build, so the existing size filter drops it on a host too small for it before this axis is consulted. The order lives in a third preference table in pkg/system, alongside the build tag and engine name tables. It is the odd one of the three: not keyed by capability, because no hardware prefers a plain build over an equivalent faster one, and matched against whole segments of a gallery ENTRY NAME rather than as a substring of a backend value. Nothing on a gallery entry declares a serving feature, and tags are not a usable substitute: gemma-4-e2b-it:sglang-mtp carries an mtp tag while ornith-1.0-9b-mtp and qwen3.6-27b-nvfp4-mtp carry none. Entry names are author-supplied free text, unlike the closed engine vocabulary, so a short marker can turn up inside an unrelated word and whole segment matching is what keeps smtp-assistant from ranking as an MTP build. The block comment over the tables now documents all three together and states what each is matched against; the ranking code names no feature, so adding one stays a one-line edit to the table. 29c49203b rejected these entries as serving configurations rather than alternative builds of the same weights. The definition is now "alternative ways to serve the same model", which includes them, so regroup 14 entries under 12 parents. Judged by the files each entry points at: the qwen3.6, qwen3.5, qwen3 and deepseek pairings are the base GGUF plus a drafter, the gemma-4 QAT MTP entries are the same QAT weights at a different quantization plus an MTP drafter, and the two sglang MTP entries describe themselves as the same model served with speculative decoding. Left separate: qwen3.6-27b-mtp-pi-tune, a finetune with its own weights, and every entry whose base model LocalAI does not ship as its own row, which is the whole Qwopus line plus gemmable-4-12b-mtp, mimo-7b-mtp:sglang and qwen3.5-4b-dflash. None of the twelve parents defines a YAML anchor, so no variants key can leak through a merge key and no empty override was needed this time. The index was edited by line insertion only. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test: check env restore errors in capability and variant specs errcheck flagged ten unchecked os.Setenv and os.Unsetenv returns in the specs added while the pre-commit hook was being skipped. Restoring an env var is exactly the place a silent failure leaks state into the next spec, so assert on it rather than suppressing the linter. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): make the mtp tag authoritative for serving-feature ranking Variant auto-selection ranks survivors by fit, then engine, then serving feature, then size. The serving-feature lookup read only whole alphanumeric segments of a variant's entry name, because tags were inconsistent: every dflash entry carried a dflash tag, but only 7 of 20 MTP entries carried an mtp tag. Tag the 13 untagged MTP entries, then teach the lookup to read tags as well as names. A tag is now the authoritative signal and is compared whole and case-insensitively, which is safe precisely because a tag is a deliberate declaration rather than free text: there is no word-inside-a-word failure mode, so the segment splitting the name half needs is unnecessary there. The name check stays as a fallback rather than being replaced. Switching to tags only would have regressed the six already-grouped entries on the day it shipped, and would depend on tagging discipline that does not exist yet. The lookup still names no feature, so adding one remains a one-line edit to servingFeaturePreferenceTokens. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): make a declared tag the sole serving-feature signal Variant auto-selection ranks survivors by fit, then engine, then serving feature, then size. The serving-feature lookup recognised a speculative build by either a declared tag or a whole segment of its entry name. Drop the name half: a tag is now the only signal. A name is author-supplied free text and a naming convention is not a contract, so reading a marker out of one infers a capability nobody declared. The gallery already had the failure in it: the four NVFP4 entries name MTP-bearing weights while setting no option that enables speculative decoding, and being live variants they were winning the feature axis without answering any faster. overrides.options was considered as the replacement and rejected. It carries spec_type:draft-mtp / spec_type:draft-dflash, which is what actually turns the feature on, but that spelling is llama.cpp's config vocabulary: ds4 spells the same feature mtp_path and sglang spells it speculative_algorithm in a referenced config. Keying a cross-backend ranking decision on one backend's option syntax would rank the other backends' builds as plain. Options are the curation-time check instead, and never reach the selection logic. With no fallback left, tag correctness is load bearing, so audit every entry against the rule "tagged when the entry configures that feature, in whatever vocabulary its backend uses". Three entries configure MTP untagged and gain the tag (hy3, glm-5.2, qwythos-9b-claude-mythos-5-1m, all spec_type:draft-mtp with no marker in their names). Four carry the tag while configuring nothing and lose it: qwen3.6-27b-nvfp4-mtp, qwen3.6-35b-a3b-nvfp4-mtp, qwopus3.6-27b-coder-mtp-nvfp4 and qwopus3.6-27b-v2-mtp-nvfp4, whose only option is use_jinja:true. The dflash side was checked independently rather than assumed consistent: all five dflash entries declare spec_type:draft-dflash and all five are tagged, so it needed no edits. Four entries keep a tag that a literal spec_type-only reading would strip, because they configure MTP through a different backend: deepseek-v4-flash-q2-mtp via ds4's mtp_path/mtp_draft, and the three sglang entries via speculative_algorithm in their referenced configs. Stripping those would contradict the reason spec_type was rejected as the signal and would demote four genuinely faster builds to plain. The index was edited by line insertion and deletion only, never round-tripped through a serializer. A resolved-tag diff across all 1272 named entries, taken after merge keys are applied, shows exactly these 7 changing and no entry gaining or losing a tag through an anchor. The two specs that pinned the name fallback are inverted rather than deleted, since a name silently promoting a build is the regression worth guarding. The whole-token guard survives on the tag path, where smtp must still not match mtp. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(gallery): make deepseek-v4-flash variant targets installable Clicking install on deepseek-v4-flash failed with "invalid gallery model". The parent entry is fine, but all four entries it was grouped with declared neither url: nor config_file:, and applyModel needs one of the two to have anything to build a config from. They carry urls: (plural), the informational HuggingFace link list, which is a different field. None of the four was ever independently installable, so grouping them routed a previously-working install into a broken entry. Give each the url: the parent already resolves through. virtual.yaml is a no-op base, and applyModel passes overrides to InstallModel separately from the fetched config, so backend: ds4, the parameters and the ssd/mtp options all still land exactly as authored. This is the same pattern the parent and many other GGUF entries in the index already use. Add the lint rule that should have caught this. checkVariantReferences only proved a target exists and is not itself a parent, which is structural validity: an entry can exist, declare no variants, and still be uninstallable. checkVariantTargetsInstallable mirrors applyModel's precondition instead, and names the parent, the target and the missing fields, because whoever hits it is reading a gallery entry and has no reason to know applyModel exists. The two index-driven resolution specs live in their own Ordered container: an Ordered container stops at its first failure, so sharing one with the lint rules let a lint breach skip them silently. Nine further entries gallery-wide have the same defect and are unrelated to variants, so they are broken installs that predate this branch. They are left alone here rather than buried in a regression fix, and widening the rule to cover every entry is deferred with them so the gate can ratchet up in one step instead of needing a skip list. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(gallery): install entries with no url or config_file on an empty base applyModel had three branches: fetch a base config from url:, build one from an inline config_file:, or fail with "invalid gallery model". An entry declaring neither is now installed on an empty base config, with overrides: and files: supplying everything. This is what the ~345 entries pointing at gallery/virtual.yaml were already getting. That stub is five lines carrying name, description and license. description and license are overwritten from the gallery entry immediately after the fetch, and the name never reaches disk because InstallModel prefers the install name. Crucially applyModel passes model.Overrides to InstallModel as a separate argument rather than merging it into the fetched config, so nothing an author writes depends on that base existing. The fetch bought a round trip to GitHub and nothing else. That makes f4ef80173 the wrong fix, so it is unwound. The four url: lines it added to the deepseek-v4-flash variants are reverted: they are a pointless network fetch now, and the family installs without them. Relaxing the branch would hide a real authoring mistake, so a payload rule replaces the base-config rule. An entry with no url, no config_file, no overrides and no files installs nothing and would leave an empty model directory while reporting success, so it is refused by name. The caller's request counts toward the payload, because its overrides and files are merged into the install exactly as the entry's own are. urls: (plural) is the informational link list and does not count, which is what the four entries that shipped broken had and why they were still uninstallable. checkVariantTargetsInstallable asserted every variant target declares a url: or a config_file:, which is no longer true and would now reject correct authoring. checkEntriesInstallSomething pins what survives instead, and covers every entry rather than only variant targets: the hazard is a half-written stanza and a parent can be one as easily as a target. The old rule was scoped to targets precisely because nine unrelated entries would have failed a gallery-wide version; those nine are valid now, so the deferred ratchet happens here in one step. 1280 entries, zero violations. Those nine (aurore-reveil_koto-small-7b-it, lfm2-1.2b, the six liquidai_lfm2 entries and deepseek-v4-pro-q2-ssd) become installable for free. Each carries overrides: and files:, and one of them is driven through the real install path in a spec. The no-fetch spec is paired rather than bare: an assertion that nothing was fetched proves nothing unless something could have been, so a control runs the same fixture with a url: pointing at a base config that is not there and asserts the install fails. Only then does the identical fixture without the url passing mean the read was skipped. Follow-up, deliberately not here: the ~345 entries still naming virtual.yaml can drop their url:. That is 345 index edits with their own risk, and mixing them in would bury this change. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * ui(models): let search bypass the variant collapse, drop the toggle The models page collapsed the gallery to one row per model by default and offered a toggle to see every individual build. Because the collapse composed with the search term, a build another entry offers as a variant could not be found by typing its name, so the toggle was the only way to reach those builds in the UI. A user who typed a name they knew existed got "no models found", which reads as "that model does not exist". Collapse is for browsing; search is for finding. An explicit search term now bypasses the collapse in the listing handler, so a name lookup returns matching entries whether or not a parent offers them. The term is trimmed once at the top of the handler, so whitespace is neither a search nor a bypass; previously an untrimmed blank term also narrowed the listing to whatever contained a space. Tag and backend deliberately do not bypass: they refine a listing the user is still reading rather than name an entry already known to exist. That makes the toggle redundant, so it goes, along with its i18n strings in all six locales, its localStorage persistence, its participation in "Clear filters" and the empty-state hint telling users to turn it off. The hint was doubly stale: it pointed at a control that no longer exists, and it was untrue exactly when a user has a search term, since searching now sees every build. The page always requests the collapsed listing. The stored preference key is left inert rather than cleaned up: nothing reads it, so a user who had the toggle off simply gets the collapsed view. collapse_variants stays on the API, off by default, because other clients want either view and the UI dropping its control is no reason to remove a working parameter. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): give the models gallery filter form a deliberate structure The filter area had accreted controls into one undifferentiated flow. The "Fits in GPU" toggle and the backend select were direct children of .filter-bar, the same wrapping container as the 18 taxonomy chips, so their position was decided by how many chips happened to wrap at the current width rather than by any layout intent. At narrow widths they were pushed past the right edge of that container's horizontal scroll and became unreachable entirely. Restructure into three bands inside the house .filter-bar-group wrapper that components/FilterBar.jsx already uses on Backends and the System tabs: 1. query scope: search plus the backend select 2. taxonomy: the chip row, alone, free to wrap 3. refinements: fits-in-GPU and context size, under a hairline rule The backend select leads the chips rather than trailing them because picking a backend disables the use cases that backend cannot serve, so it gates the row below it. Fits-in-GPU and context size share a band because they are one control group: the context size is the length the VRAM estimate is computed at, and that estimate is what the fits filter tests against. Chips had no visible keyboard focus indicator. The global focus ring is wrapped in :where(), so it carries the specificity of a bare :focus-visible, ties with .filter-btn and loses on source order, leaving focused chips showing their resting drop shadow. Restate the ring where it outranks both resting and hover. Also: aria-pressed on the chips, a real label association and aria-valuetext on the context slider (it steps over an index, so it announced "2"), disabled chip styling moved off inline styles, a prefers-reduced-motion block for the chip transition, and the hard-coded English "Context:" moved into all seven locales. No behaviour change: same filters, same state, same requests. Page reset on change, localStorage persistence and "Clear filters" verified unchanged. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): let the models recommendations panel fade into the background The "Recommended for your hardware" strip rendered at full height on every visit regardless of how many models were already installed, costing 186px at 1600px wide (287px at 1100px, where its cards wrapped to two rows) and pushing the first gallery row to y=554 / y=703. Make its prominence track how much the user still needs it. The panel now defaults to a one-line summary once anything is installed, and both the collapse choice and the existing dismissal persist: collapsed = explicit user choice, if one exists : installedCount > 0 The preference is three-valued on purpose. A boolean cannot tell "the user expanded it" apart from "the user has never chosen", and those need opposite handling when the installed count later crosses zero: someone who deliberately opened the panel on an empty instance should not have it collapse out from under them when their first model finishes installing. Collapsed keeps the card, icon, title and a suggestion count, so the panel is recovered by clicking what you are already looking at rather than by hunting. Expanded is unchanged, because for a user with nothing installed it was never the problem. Collapsed reclaims 145px at 1600 and 420, and 246px at 1100. Models.jsx gains a statsLoaded flag: stats initializes to installed:0, so reading it before the fetch resolves would render expanded and collapse a frame later, which is exactly the layout shove this removes. The dismissal key moves to the page's localai-models-* convention; the old localai_rec_models_dismissed is still read, never written, so an existing dismissal is honoured rather than resurrected by the rename. Accessibility: the disclosure is a real button whose accessible name is the visible title alone, with state on aria-expanded and aria-controls resolving in both states, because the grid is hidden via the hidden attribute rather than unmounted. That also keeps the four install buttons out of the tab order while collapsed. The app's global focus ring applies; no per-component outline is added, per the warning in App.css. Reveal animates opacity and transform only, never height, and both it and the chevron rotation are disabled under prefers-reduced-motion. Only en had a recommended block, so the other six locales were falling back to English for the whole panel. Translated the complete block rather than adding one orphaned key to files that would still render the title in English. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(downloader): recover from a leftover .partial on non-HTTP URIs An interrupted download leaves a `<file>.partial` behind. The partial handling in DownloadFileWithContext gated resume on `err == nil && uri.LooksLikeHTTPURL()`, so for any URI that is not literally http(s) the branch fell through to `else if !errors.Is(err, os.ErrNotExist)`, which with a nil err is true. The download then failed with an error wrapping nil: failed to check file ".../Ternary-Bonsai-27B-Q2_g64.gguf" existence: <nil> Every gallery file URI uses `huggingface://`, so a single interrupted download made that model permanently uninstallable until someone deleted the partial by hand. The `<nil>` in the message compounded it by pointing debugging at a filesystem failure that never happened. Restructure the handling as an explicit switch over the four real states: partial exists and is resumable, partial exists and is not resumable (discard and restart, as already done for an HTTP server without range support), no partial, and a genuine stat failure. The error branch is now only reachable with a non-nil error, names the path that was actually stat'd, and wraps with %w. Discarding is required for correctness and not merely convenience: the writer opens the partial with O_APPEND, so an un-resumed download would concatenate a fresh body onto stale bytes. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): tell the models gallery's variant rows apart, and let browsing see every build Both variant surfaces rendered name, backend and size. For two builds of one model that is close to no information: a variant exists precisely because the same weights are offered another way, so the backend usually matches and the sizes usually land within a few hundred megabytes. Comparing ternary-bonsai-27b-pq2 against ternary-bonsai-27b-q2-g64 meant reading two names that differ by a suffix nobody has defined anywhere in the UI. Report the quantization and the serving features on VariantView, and derive both server-side from the referenced entry rather than parsing names in the browser, so every client reads the same format out of the same file the installer will hand the backend. Quantization comes from overrides.parameters.model first, falling back to the file list. That order is load bearing: entries routinely ship a vision tower alongside the language model at a different quantization, so reading the file list first reports the mmproj's format. Matching walks `-` and `.` delimited segments right to left; `_` deliberately does not split, because it separates the parts INSIDE a quant token and splitting on it reports Q4 for a Q4_K_M build. A second, looser pass takes a segment's `_`-delimited tail, which catches the gemma-4-E2B_q4_0-it.gguf style; it runs second so a precise match can never lose to a fuzzy one further right in the name. An entry naming no format reports nothing, which is the honest answer for a backend served from a directory of weights. Features are the same tag-against-vocabulary match servingFeatureRank already ranks on, over the same host preference list. A build can therefore never be shown as faster than one selection did not actually reward, nor rewarded without being shown; a spec pins that agreement rather than trusting it. The compact dropdown gets the quantization on its meta line and the bare feature token. The detail row, which has the room, gets the quantization as its own monospaced column so precision lines up down the list, and the feature spelled out, because DFLASH names nothing to a user who has not met it. The referenced entry's description stays out of both: the detail row already renders the parent's prose above the table, and a second block per variant would push a three-variant list past a screen to restate what the columns now say precisely. The collapse toggle comes back. 462583f38 dropped it once search bypassed the collapse, on the reasoning that nothing was unreachable any more. That holds for finding a build whose name you know and does not hold for browsing: no sequence of actions enumerated the 68 builds the default view hides. Collapse is for browsing and search is for finding, and the toggle was the browsing half. It goes in the refinements band 0d4823362 established, not back among the taxonomy chips where its position depended on how many chips happened to wrap. It leads that band because it decides how many rows the other two refine over, and because unlike fits-in-GPU it is unconditional: a host with no GPU still browses. The search bypass is untouched and re-checked by a spec in the toggle's default state, since restoring the control must not restore the dead end it replaced. The empty-state hint returns but only without a search term, because a term bypasses the collapse and the hint would otherwise point at a control that cannot change the result. The stored preference reads 'on'/'off' only: an older build wrote '1'/'0' from an effect that ran on mount, so those record that the page was opened, not that anyone chose a view. Also fixes a latent flake it exposed. The collapse_variants spec compared whole response bodies byte for byte, and the listing envelope carries live host telemetry that drifts between two calls milliseconds apart, so it was asserting on the machine's memory pressure. It now compares everything the parameter governs -- the entries, their serialization and the paging -- and is green 25/25 where it was failing about one run in three. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ui): let the models gallery show a variant's full details The variant list in an entry's expanded detail row says how the builds differ: name, backend, quantization, size, and the auto-selected, base and serving-feature markers. It cannot say what any one of them is. A variant's own description, tags, license, source links and file list are unreachable anywhere in the UI, because while the collapse is on a variant has no gallery row of its own. Give each variant row an info control that reveals its entry, rendered by the same ModelDetail a top-level row gets, so a field added to the detail view appears here too. variantData is withheld from the nested render: a variant may declare variants of its own, and recursing would nest a picker inside a picker two levels deep already. An inline disclosure rather than a modal. The control sits inside a table row that is already expanded, inside a variant list within that; a dialog opened from there stacks a dismissal on a dismissal for a handful of extra fields about the entry the user is already reading, and breaks the page's own expand idiom. The third level is carried by an inset and a left rule instead of another card. The entry is fetched by exact name from the listing, once, on first use. The listing already returns every field the detail view renders, and a search term bypasses the variant collapse server-side, so no new endpoint is needed and neither the listing nor DescribeVariants gains any work. Expanding a row costs nothing; a variant nobody opens costs nothing. A name the listing no longer returns is stated, not blanked: an empty panel reads as a rendering fault rather than as a lookup that came back empty. The control is a sibling of the install button, not a descendant, so asking about a build can never install it. The variant list keeps its content-sized columns via a trailing filler track instead of max-content sizing, so the rows are unchanged while the panel spanning them gets the pane width its file table needs. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(gallery): let search respect the collapse instead of switching it off The models listing collapsed to one row per model, and an explicit search term turned that off wholesale. Searching while collapsed therefore answered with the individual builds a parent already offers, which are exactly the rows the view the user asked for has no place for: typing "mtp" returned qwen3.6-27b-nvfp4-mtp, a row that is invisible the moment the box is cleared. The bypass was the right shape of fix for the wrong half of the problem. What a search must not do is answer "no models found" for a build the gallery does hold; that does not require abandoning the grouping the user asked for. So the term is now matched against every entry either way, hidden builds included, and the collapse decides how a match is reported rather than which matches exist. Collapsing stops being a filter that drops rows and becomes a substitution: a match on a build another entry offers is reported as that entry, the one installable in its own right. Nothing becomes unfindable and nothing comes back that the requested view cannot show. Substitution happens after search, tag and backend, so every filter is judged against the build that really carries the name, tag or backend rather than against a parent that merely offers it; the other order would let backend=vllm match a parent whose own backend is something else. The price is that the surfaced row shows the parent's own metadata while the match was on a variant, which is what grouping means, and the alternative is claiming the gallery holds no such build. It happens before the count and the page math, so both describe the rows actually handed out rather than the matches that produced them. A parent already in the result keeps its own position and absorbs its matching variants there, which is what leaves the browsing listing ordered exactly as it was; a parent surfaced only by a variant takes the position of the first variant that surfaced it. Either way it appears once, however many of its builds matched and whether or not it matched itself. Search preserves gallery order rather than scoring, so a surfaced parent has a real position rather than an invented one. VariantParents never reports an entry that declares variants of its own, so a parent is never itself hidden and one hop always lands on a visible row. The handler follows exactly one anyway: refusing the second is what makes a gallery the linter would have rejected terminate rather than loop. The empty-state hint pointing at the toggle goes with it for every server-side filter. Substitution means a match is always reported as some row, so the collapse can no longer be why a term, a chip or a backend came back empty, and naming it there sends the user to a control that cannot change the result. It survives for the fits filter alone, which runs in the browser after the substitution and judges the surfaced entry's own size: there the build that fits really can be filtered out along with a parent that does not. Searching a build's exact name while collapsed now answers with its parent, so the result no longer contains the string the user typed. That is intended, and the row is the one they can act on, but it is a real rough edge: nothing on the row explains the connection. Closing it properly means reporting which variant matched so the UI can say so, which the listing does not do today. ResetGalleryModelCache is added for tests. The model cache is a package global keyed by nothing, so a background refresh one spec triggers can land in the middle of the next and answer it with the previous spec's gallery; the extra specs here made that fail about one run in five. It waits for the in-flight refresh to publish before clearing, since clearing alone only narrows the window. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-07-20 18:43:02 +02:00
LocalAI [bot]andEttore Di Giacinto	b00422e45f	feat(backends): add LongCat video and avatar generation (#10792) ... * feat(backends): add LongCat video and avatar generation Assisted-by: Codex:GPT-5 [apply_patch] [exec_command] [web] * refactor(config): declare model I/O modalities Make model configs declare input and output modalities so capability discovery no longer branches on backend or checkpoint names. Complete the LongCat gallery and user documentation, make the SDPA patch apply to the pinned upstream revision, and stabilize the Agent Jobs race exposed by the required hook. Assisted-by: Codex:GPT-5 [web] --------- Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-07-12 23:58:46 +02:00
LocalAI [bot]andEttore Di Giacinto	40dae953f4	feat: interleaved thinking with tool calls (reasoning_content alias + Anthropic thinking blocks) (#10744) ... * feat(schema): accept reasoning_content as inbound alias for reasoning Interleaved-thinking clients (cogito, vLLM/DeepSeek-style) emit reasoning_content on assistant turns. Accept it as an inbound alias so reasoning survives the tool-result loop; canonical reasoning wins when both are present. Emission is unchanged (still reasoning). Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(schema): pin interleaved reasoning+tool_calls round-trip Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * test(openai): pin reachedTokenBudget truncation detection Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(anthropic): add thinking and signature fields to content blocks Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(anthropic): parse inbound thinking blocks into reasoning Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(anthropic): emit thinking blocks with synthetic signature on tool turns Extract buildAnthropicContentBlocks so non-streaming content assembly is unit-testable, and prepend a thinking block (with an opaque synthetic signature) before text/tool_use blocks when the request opts into thinking. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(anthropic): stream thinking_delta and signature_delta before tool_use Extract anthropicStreamSequence so the streaming block order is unit-testable, and emit content_block_start(thinking) -> thinking_delta -> signature_delta -> content_block_stop before the tool_use block sequence when thinking is enabled. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs: add interleaved thinking with tool calls guide Add a features guide describing interleaved thinking: an assistant turn carrying reasoning and tool_calls together, the reasoning-round-trip contract (including the reasoning_content inbound alias and Anthropic thinking blocks with a synthetic signature), per-backend enablement (reasoning_format for llama.cpp, reasoning_parser/tool_call_parser for vLLM/SGLang plus the vLLM auto-config hook), a worked request/response example, and known limitations. Cross-link from model-configuration, text-generation, and openai-functions. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-07-08 16:45:43 +00:00
LocalAI [bot]andEttore Di Giacinto	b0959d4756	feat(api): add GET /v1/models/capabilities endpoint (#10687) ... Additive superset of /v1/models that enriches each model entry with the capabilities it supports plus its input/output modalities (text / image / audio / video). Clients that only understand /v1/models are unaffected -- they simply never call the new route. Audio and video *input* are derived from the model's multimodal limits (vLLM limit_mm_per_prompt), which no single usecase FLAG expresses. That gap is exactly why a plain capability list is insufficient and this enriched endpoint exists: an attachment router can now decide whether an image/audio/video file can go to the active model directly, or must be converted/transcribed first. Capability derivation lives in core/config as the single source of truth (ModelConfig.Capabilities / InputModalities / OutputModalities / VisionSupported / ...); the Ollama capability surface now delegates to it instead of keeping a parallel copy. Vision is gated on chat/completion capability so a MediaMarker hydrated onto a non-chat model (e.g. a pure ASR/TTS backend) no longer reports a false vision capability. Read-only listing: no new FLAG_* flag, reuses the existing `models` swagger tag, and intentionally exposes no MCP admin tool (there is nothing to manage conversationally). Assisted-by: Claude:claude-opus-4-8 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-07-05 08:51:55 +02:00
Richard Palethorpe	eb32cd9073	feat(realtime): eager blocking pipeline warm-up + /backend/load API (#10662) ... Realtime sessions previously lazy-loaded each pipeline sub-model (VAD, transcription, LLM, TTS) on first use, so every cold session paid a per-request model-load stall and load errors only surfaced mid-stream. Warm the whole pipeline eagerly and blockingly at session start (including the voice-gate speaker-recognition model, which an enforced gate blocks each utterance on; compaction's summary_model stays lazy since it only runs off the response path): - Add backend.PreloadModel / PreloadModelByName as the single load path for every modality (no transcription special-case; backend-omitted configs are deprecated). - The realtime session blocks on Model.Warmup and returns a model_load_error to the client if any stage fails to load; updateSession warms in the background. Opt out per pipeline with pipeline.disable_warmup, exposed as a UI toggle via the config-metadata registry. Add a LocalAI-native POST /backend/load (and /v1/backend/load) that pre-loads a model -- expanding realtime pipelines into their sub-models -- as the inverse of /backend/shutdown. There is one preload engine (backend.PreloadStages): the realtime Warmup methods, /backend/load and the --load-to-memory startup flag all use it, so --load-to-memory now also expands pipeline models and records load-failure traces. Pipeline sub-model alias resolution is likewise shared (ModelConfigLoader.LoadResolvedModelConfig). Surface the endpoint everywhere an admin manages models: - MCP admin tool load_model (httpapi + inproc clients, safety/catalog prompts, catalog/dispatch tests). - "Load into memory" action in the React models UI. - Swagger regenerated; docs moved to the general backend-monitor page since it is not realtime-specific. Fix a Traces UI crash ("json: unsupported value: -Inf"): audio-snippet RMS/peak now floor at a finite dBFS, and backend-trace data is sanitized to drop non-finite floats before marshaling. The sanitizer is copy-on-write -- it runs on every RecordBackendTrace, so containers are only re-allocated on the paths that actually changed. Migrate core/http/openresponses_test.go onto the prebuilt mock-backend the rest of the http suite already uses -- it was the last spec still pointing at a real HuggingFace model, so it 404'd wherever no vision backend was built -- and fix its item_reference specs to send the spec's "id" field instead of "item_id", which the handler never accepted. Assisted-by: Claude:claude-opus-4-8 Claude Code Signed-off-by: Richard Palethorpe <io@richiejp.com>	2026-07-03 18:00:37 +02:00
Richard Palethorpe	5d0c43ec6e	feat(realtime): Semantic VAD EOU token (#10444) ... * feat(realtime): EOU-driven semantic_vad turn detection Add a `semantic_vad` turn-detection mode to the realtime API that feeds the transcription model live and decides "the user finished speaking" from the `<EOU>` end-of-utterance token rather than from silence alone. When EOU fires the turn commits immediately (~0.3s); otherwise it falls back to an eagerness-scaled silence threshold (low/med/high = 8/4/2s). Plumbing, bottom to top: - proto: `AudioTranscriptionLive` bidirectional RPC (config-first oneof, mono float PCM @16k, ready-ack / Unimplemented degrade signal) plus `TranscriptResult.eou` for the unary retranscribe gate. - pkg/grpc: client/server/base/embed scaffolding for the bidi stream, modeled on AudioTransformStream; release stream conns on terminal Recv. - parakeet-cpp: live transcription RPC with per-C-call engine locking (one live stream per turn, finalize+free at commit); bump parakeet.cpp to ABI v5 — incremental StreamingMel (no more quadratic per-feed mel recompute that delayed EOU on long turns) and the <EOU>/<EOB> split; strip the literal <EOU>/<EOB> from offline text and set Eou. - core/backend: LiveTranscriptionSession wrapper + pipeline `turn_detection:` config block (type/eagerness/retranscribe). - realtime: semantic_vad integration — live input captions streamed as transcription deltas while the user speaks, EOU-immediate commit with eagerness fallback, optional retranscribe gate (batch re-decode must also end in <EOU> to confirm), clause synthesis off the LLM token callback, and per-turn live-transcription / model_load telemetry. - UI: show the realtime pipeline components as a vertical list. Docs and tests included; opt-in via the pipeline YAML or per-session `session.update`. Non-streaming STT backends degrade to silence-only. Assisted-by: Claude Code:claude-opus-4-8 [Read] [Edit] [Write] [Bash] Assisted-by: Claude Code:claude-fable-5 [Read] [Edit] [Bash] Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(realtime): explicit formally-verified state machines + parakeet streaming driver The realtime API had several implicit state machines whose state was inferred from scattered booleans, channels, and five separate mutexes, leaving illegal/inconsistent states reachable. Make them explicit and keep the implementation in step with a formal design; rework the parakeet streaming backend along the same lines. Realtime state machines (M1-M5). Each is a sealed sum-type State/Event/Effect with a total, pure Next(state,event)->(state,[]effect) behind a single-writer Coordinator: M1 conncoord connection lifecycle: VAD toggle + once-only teardown (replaces vadServerStarted + a `done` channel closed from two sites). M2 turncoord turn detection: collapses speechStarted and the live-stream "turn open" flag into one state, so discardTurn can no longer desync them and suppress the next onset. M3 respcoord response coordination: serializes the dual-writer start/cancel so at most one response is live; one response.done per response.create. M4 compactcoord conversation compaction: single-flight (replaces the `compacting atomic.Bool` CAS). M5 ttscoord TTS pipeline: open->closing->closed, idempotent wait(), rejects enqueue-after-close (was a silent drop). The Coordinator/Sink/Next plumbing — only the sealed types and Next differed per machine — is extracted once into core/http/endpoints/openai/coordinator as a generic Coordinator[S,E,F]; each machine keeps its public API via type aliases, so no sink, call-site, or test moved. Hierarchy. session_lifecycle.fizz models M1 as the parent region with its children (M2/M3/M4) as one statechart and asserts ChildrenDieWithParent (conn torn => all children terminal, none start after teardown). respcoord and compactcoord gain an absorbing Terminated state + Shutdown event; conncoord's teardown drives the children terminal. This closes a compaction teardown gap: a fire-and-forget compaction could outlive a torn session — compactionSink now takes a session-scoped cancellable context + WaitGroup and joins the in-flight summarize+evict on shutdown. Formal verification. formal-verification/ holds one authoritative FizzBee spec per machine plus the composition spec, each with an always-assertion and a documented one-line edit that makes the checker fail (verified non-vacuous). scripts/realtime-conformance.sh is fail-closed: all Go conformance suites under -race AND a model-check of every .fizz spec; a missing FizzBee is a hard error (only the loud REALTIME_CONFORMANCE_SKIP_FIZZBEE=1 bypasses it, never in CI). FizzBee is pinned by sha256 and installed via scripts/install-fizzbee.sh into .tools/ (gitignored). Wired as make test-realtime-conformance, a CI workflow, and a pre-commit path filter. Go conformance tests are Ginkgo/Gomega (per the repo's forbidigo lint): transition tables + fixed-seed property walks + concurrent/-race specs, no rapid dependency. Design map: docs/design/realtime-state-machines.md. Parakeet streaming backend. The same treatment applied to the parakeet-cpp streaming paths: - AudioTranscriptionStream returns codes.Unimplemented for non-streaming models instead of decoding offline and emitting it as one delta + final. A client that asked for streaming learns the model cannot stream rather than receiving a batch result shaped like a stream. New grpcerrors.StreamTranscriptionUnsupported carries that signal; the HTTP /v1/audio/transcriptions stream path surfaces it as an SSE error event. Mirrors AudioTranscriptionLive, which already did this. - utteranceBoundary (boundary.go): a single definition of the end-of-utterance latch, replacing three open-coded finalEou toggles. Modelled as a two-valued type so illegal states are unrepresentable. - Shared decode driver (driver.go): streamFeedResult (one per-feed event) + feedChunk (hides the ABI v4 JSON vs text-only split) + feedSlices + flushTail. The feed loop is written once. - AudioTranscriptionLive becomes a bidi adapter: it streams the per-feed {delta,eou,eob,words} the realtime turn detector consumes and a terminal FinalResult carrying only Text. Segments/duration/eou are offline-only and no longer produced (nor read) on the live path; liveTraceState drops the terminal eou and keeps the per-feed eou_events count. - AudioTranscriptionStream + streamJSON merge into one driver-based function; streamSegmenter is generalized to the unified event with a text-only fallback that preserves the legacy (no-words) library's per-utterance segmentation. Verified: build/vet/gofumpt clean, golangci-lint 0 issues, all coordinator and parakeet packages under -race, the fail-closed conformance gate green, and make test-realtime (12 e2e WS+WebRTC). Assisted-by: Claude:claude-opus-4-8 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> --------- Signed-off-by: Richard Palethorpe <io@richiejp.com>	2026-06-30 09:01:22 +02:00
LocalAI [bot]andEttore Di Giacinto	f0d0bff232	fix(llama-cpp): stop reinterpreting plain-string message content as JSON (#10524) (#10538) ... The llama-cpp gRPC backend reconstructs OpenAI messages from proto for the tokenizer-template path and blindly json::parse'd each message's content string. LocalAI's Go layer always flattens content to a plain string, so a user prompt that merely looks like JSON (e.g. mealie's ingredient array ["1/4 cup brown sugar", ...]) was reinterpreted as structured content parts and rejected by oaicompat_chat_params_parse with "unsupported content[].type". Normalize content per role instead: user/system/developer content is opaque text and is never JSON-sniffed; assistant/tool content still collapses a literal JSON null/object (tool-call bookkeeping) to a string, but a plain string is never turned into an array/scalar. The array defense is role-independent, so the role gate only governs the benign null/object case. While here, extract the duplicated per-message reconstruction and the pre-template content sanitization into shared, unit-tested helpers (message_content.h) so the streaming (PredictStream) and non-streaming (Predict) paths cannot drift. This removes ~490 lines of copy-pasted defensive code, the dead tool-role parse branches, and the redundant Predict-only tool_calls branch, while preserving the prior #7324 (null content -> "") and #7528 (tool array content -> string) fixes. Tests: - backend/cpp/llama-cpp/message_content_test.cpp: standalone C++ unit tests for all three helpers (#10524, #7324, #7528, multimodal), discovered and run by `make test-backend-cpp` and a new generic tests-backend-cpp CI job. Also wired as an opt-in CMake/ctest target (-DLLAMA_GRPC_BUILD_TESTS=ON). - core/schema/message_test.go: Go regression pinning that ToProto flattens a JSON-array-looking text part to the verbatim string. - prepare.sh now copies message_content.h into the build tree. Assisted-by: Claude:claude-opus-4-8 [Claude Code] Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-06-27 01:42:05 +02:00
LocalAI [bot]andEttore Di Giacinto	600dafd20b	feat(ced): sound-event classification backend (CED audio tagger) (#10425) ... * feat(ced): sketch sound-classification backend (CED audio tagger) Wires ced.cpp (CED, 527-class AudioSet sound-event tagger; baby cry, footsteps, glass, alarms, dog bark) into LocalAI as a Go/purego backend. SKETCH (backend skeleton real; core REST wiring + CI/gallery is a checklist in DESIGN.md): - backend/backend.proto: new SoundDetection rpc + SoundClass messages (run `make protogen-go` to regenerate pkg/grpc/proto). - backend/go/ced: main.go (purego dlopen libced.so + ced_capi.h), goced.go (Ced gRPC backend: Load + SoundDetection), Makefile (clone-at-pin CED_VERSION, ggml static-PIC shared build), run.sh, package.sh, .gitignore. - DESIGN.md: REST /v1/audio/classification wiring (handler/route/capability registration checklist), gallery/index + CI registration, and a scoping note for the realtime/websocket live-recognition path (sliding-window classify over the existing ws transport + voicegate; the ced C-API per-PCM entry point is already window-friendly). Backend code does not compile until protogen-go regenerates the pb types and a libced.so is built (Makefile clones+builds it). Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ced): REST /v1/audio/classification endpoint + capability registration Wires the ced sound-event classification backend (AudioSet audio tagger) end to end through the REST surface, mirroring the transcription path. - Handler: core/http/endpoints/openai/sound_classification.go parses the multipart audio upload, temp-files it, resolves the model config and calls the SoundDetection RPC; returns {model, detections[]} JSON. - Backend wrapper: core/backend/sound_classification.go (ModelSoundDetection) loads the model and normalizes the proto response into schema types. - Schema: core/schema/sound_classification.go (SoundClassificationResult). - gRPC layer: SoundDetection wired through the LocalAI wrapper (interface, Backend client, Client, embed, server, base default) so the loader-typed client exposes the RPC; proto regenerated via make protogen-go. - Route: POST /v1/audio/classification (+ /audio/classification alias) with the audio/multipart default-model middleware in routes/openai.go. - Capability surfaces: swagger @Tags/@Router on the handler; FLAG_SOUND_ CLASSIFICATION usecase flag + UsecaseSoundClassification + UsecaseInfoMap + GuessUsecases + ModalityGroups + GetAllModelConfigUsecases; meta usecase option; /api/instructions audio area updated; auth RouteFeatureRegistry + FeatureAudioClassification (APIFeatures, default ON) + FeatureMetas; UI usecaseFilters, capabilities.js CAP_SOUND_CLASSIFICATION, Models.jsx filter + i18n; docs page features/audio-classification.md + whats-new + crosslink. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ced): realtime sound-event detection over the websocket API When a realtime pipeline configures a sound-classification model, each VAD-committed utterance (the same window the transcription path produces) is also run through the CED sound-event classifier and the scored AudioSet tags are emitted as a new server event. No new backend rpc is needed: the SoundDetection gRPC method already exists on this branch. - config: add Pipeline.SoundDetection (yaml/json sound_detection,omitempty) beside Transcription/VAD. - realtime: add Model.SoundDetection(ctx, audio, topK, threshold) to the ModelInterface; implement it on wrappedModel and transcriptOnlyModel by calling backend.ModelSoundDetection with the session's sound-classification model config (mirrors how Transcribe dispatches). Load the optional config in newModel / newTranscriptionOnlyModel; nil config keeps it additive. - types: add ConversationItemSoundDetectionEvent (item_id, content_index, detections[]{label,score,index}) with type conversation.item.sound_detection, its ServerEventType constant and MarshalJSON, mirroring the transcription completed event. - realtime: add emitSoundDetection (unary path: classify the committed window, build the event, t.SendEvent) and wire it at the utterance-commit hook right after emitTranscription; gated on session.SoundDetectionEnabled (resolved from Pipeline.SoundDetection at session setup, defaults top_k=5, threshold=0). Its error is logged via xlog but never aborts the turn. - test: Ginkgo specs for emitSoundDetection (tags emitted, empty detections, classifier error) plus a SoundDetection method on the fakeModel double. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ced): implement SoundDetection in nodes backend test doubles The SoundDetection method added to the grpc backend interface left two test doubles (fakeBackendClient, fakeGRPCBackend) incomplete, so core/services/nodes failed to compile under `go vet`/`go test` (go build missed it: the doubles live in _test.go). Add the method to both, mirroring their existing Detect mock. Repairs CI for the nodes package. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ced): decouple realtime sound detection from VAD (sound-only sessions) Sound-event detection must activate on sounds, not speech, so it no longer runs through the voice VAD/transcription path. A sound-detection-only pipeline (sound_detection set, no transcription/LLM) now: - is accepted by prepareRealtimeConfig (sound_detection counts as a pipeline stage), - builds a lightweight model via newSoundDetectionOnlyModel (no VAD/STT/LLM/TTS loaded), and - defaults the session to turn_detection none (no VAD) with no transcription stage, so the client drives windowing via input_audio_buffer.commit (option A: client-side sliding window). The per-PCM C-API already supports arbitrary windows. commitUtterance gains a sound-only branch: it emits the conversation.item.sound_detection event (scored AudioSet tags) and stops - no transcription, no LLM response. generateResponse is now guarded on a transcription stage being present, so a sound-only turn never invokes the LLM. Existing transcription/VAD sessions are unchanged (additive). Added a commitUtterance sound-only Ginkgo spec asserting it emits the sound event and neither transcribes nor generates a response. go vet + golangci-lint (new-from-merge-base) clean; openai suite green. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ced): register sound-classification backend in gallery + CI Mechanical backend-image registration for the ced sound-event classifier, mirroring the parakeet-cpp Go/purego backend everywhere it is wired up. - .github/backend-matrix.yml: add the ced build matrix, field-for-field copies of the parakeet-cpp entries (cpu amd64/arm64, cublas cuda 12/13 amd64, l4t cuda-13 arm64, l4t-jetpack cuda-12 arm64, sycl f32/f16, vulkan amd64/arm64, rocm hipblas, and the metal darwin entry), changing only backend and tag-suffix. dockerfile stays ./backend/Dockerfile.golang. - backend/index.yaml: add the &ced meta anchor (capabilities map per platform) plus ced-development and the per-arch image entries, each uri/mirror tag-suffix matching the matrix exactly. The model gallery (GGUF) entry is intentionally deferred pending the HuggingFace publish (TODO note inline). - scripts/changed-backends.js: add an explicit item.backend === "ced" branch in inferBackendPath mapping to backend/go/ced/, same mechanism and ordering as the parakeet-cpp branch (before the generic golang fallthrough). - .github/workflows/bump_deps.yaml: register mudler/ced.cpp -> CED_VERSION in backend/go/ced/Makefile so the daily bot bumps the pin. - swagger/{docs.go,swagger.json,swagger.yaml}: regenerated via make swagger so the existing /v1/audio/classification annotations land in the generated spec. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ced): server-side windowing for realtime sound detection (option B) Adds an optional server-driven sliding-window classifier so a sound-only realtime client only has to stream audio (no input_audio_buffer.commit): - Pipeline.sound_detection_window_ms / sound_detection_hop_ms config knobs. When both > 0 on a sound-only session, the server classifies the last window of streamed audio every hop and emits a conversation.item.sound_ detection event; the input buffer is trimmed to one window so a long stream stays bounded. When unset, the session stays client-driven (option A). Runs independent of VAD (sound events are not speech). - handleSoundWindow (ticker) + classifySoundWindow (one tick, extracted so it is unit-testable) + writeWindowWAV, which declares the true InputSampleRate (NewWAVHeaderWithRate) so the classifier resamples correctly. Goroutine is started after toggleVAD and torn down with the session (close + wg.Wait). - Register pipeline.sound_detection (+window_ms/hop_ms) in the config meta registry; the earlier realtime commit added pipeline.sound_detection without a registry entry, failing TestAllFieldsHaveRegistryEntries. This fixes that and covers the two new knobs. Tests: classifySoundWindow emits an event + trims the buffer to one window, no-ops on too-little audio; writeWindowWAV declares the given sample rate. go build/vet + golangci-lint (new-from-merge-base) clean; config + openai suites green. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ced): add ced-base GGUF model gallery entries (f16 + q8_0) The ced-base weights are now published at mudler/ced-base-gguf (Apache-2.0, converted from mispeech/ced-base). Adds gallery/ced.yaml (backend: ced + known_usecases: sound_classification) and two gallery/index.yaml entries (ced-base-f16 default, ced-base-q8 smallest) with sha256-pinned files, and removes the now-resolved TODO from backend/index.yaml. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(ced): add tiny/mini/small GGUF model gallery entries Publishes the rest of the CED family (same architecture, metadata-driven port verified end-to-end on ced-tiny) to mudler/ced-{tiny,mini,small}-gguf and adds their f16 + q8_0 gallery entries: ced-tiny (5.5M, edge/Pi-class) f16 11MB / q8_0 6MB ced-mini (9.6M) f16 19MB / q8_0 11MB ced-small (22M) f16 42MB / q8_0 23MB All sha256-pinned. ced-base remains the accuracy default. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore(ced): point gallery entries at the consolidated mudler/ced-gguf repo All CED quantizations (tiny/mini/small/base, f16/q8_0) now live in a single HuggingFace repo, mudler/ced-gguf, instead of per-model repos. Repoint the 8 gallery model entries' urls + file uris accordingly. sha256 and filenames are unchanged. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore(ced): bump CED_VERSION to the short-clip fix Pin the ced backend to ced.cpp 99c6ed3, which fixes a crash on any clip shorter than target_length (~10.11s): time_pos_embed was added at its full 63-frame grid instead of being sliced to the clip's actual time grid, tripping ggml_can_repeat in ggml_add. Surfaced by the live realtime e2e (sub-10s windows) and gated with a short-clip parity test upstream. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * docs(ced): list ced.cpp as a LocalAI-team engine + backend-guide directive - README.md: add ced.cpp to the "native C/C++/GGML engines developed and maintained by the LocalAI project" table. - docs/content/features/backends.md: add a Sound Classification backend category (sound-event classification / audio tagging) listing ced.cpp. - .agents/adding-backends.md: add a "Documenting the backend" section and two verification-checklist items requiring new backends to be documented in the backends.md category list, and in-house native engines to be added to the README maintained-engines table. This directive was missing. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * chore(ced): repin CED_VERSION to the v0.1.0 release commit ced.cpp history was squashed into a single release commit (tagged v0.1.0), so the previous pin (99c6ed3) no longer exists upstream. Pin to c04ac14, the v0.1.0 release commit, so the backend builds against a commit that exists. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(ced): silence gosec G304/G103 + govet unsafeptr on audited paths - sound_classification.go: os.Create(dst) where dst = temp dir + path.Base of the upload (no traversal). #nosec G304, matching the depth-anything-cpp handler. - goced.go: reading a NUL-terminated C string from a libced-owned buffer. #nosec G103 (gosec) + //nolint:govet (golangci-lint's unsafeptr check), since the uintptr is a C-owned malloc'd buffer, not Go-GC memory. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-06-22 01:00:28 +02:00
Richard Palethorpe	3fa7b2955c	feat(pii): NER tier engine — privacy-filter.cpp backend + NER-centric PII filter (#10360) ... Squashed feat/pii-ner-tier-engine rebased onto master (was 45 commits; see backup/pii-ner-tier-engine-prerebase). Net change: - privacy-filter.cpp: standalone GGML engine for the openai-privacy-filter PII/NER token classifier, wired as a LocalAI gRPC backend (CPU/CUDA/Vulkan). TokenClassify moves off the patched llama.cpp path onto this backend. - PII filter reworked to be NER-centric (encoder/NER detection tier scanning whole conversations as one document), with a recreated bounded restricted- regex secret-matching pattern detector tier alongside it (per-model pii_detection.builtins / .patterns + core/services/routing/piipattern). - Detection labelled by source (ner vs pattern); backend trace / confidence / debug observability; analyze/redact exposed as a synchronous API. - Instance-wide default detector policy + per-usecase default-on; request filtering extended to completions, embeddings, edits & Ollama. - React UI: NER-centric PII editor, detector-models table, pattern/builtins editor, middleware default-policy UI. - Gallery: privacy-filter-multilingual token-classify model + NER install filter; token_classify known_usecase; batch sized to context for NER models. privacy-filter backend registered in the backend gallery (cpu/vulkan/cuda-13 meta + image entries with a capabilities map) matching its CI matrix jobs, and an /import-model auto-detect importer (PrivacyFilterImporter, narrow privacy-filter GGUF detection) replacing the prior pref-only registration. Reconciled against master's independent evolution: - Dropped master's PIIPatternOverrides feature (global-pattern runtime overrides + /api/pii/patterns API + runtime_settings.json persistence). The per-model NER + pattern-detector design supersedes it; it was built on the global redactor pattern set this branch replaced. - Reverted the llama.cpp Score carry-patch (0006-server-task-type-score): removed the patch and restored master's grpc-server.cpp Score RPC (direct llama_decode, slot-loop bypass) and LLAMA_VERSION pin, plus master's model_config validation forbidding score + chat/completion/embeddings on llama-cpp. token_classify is unaffected (it runs on the privacy-filter backend, not llama-cpp). Assisted-by: Claude:claude-opus-4-8 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com>	2026-06-18 11:45:22 +01:00
LocalAI [bot]andEttore Di Giacinto	294170d3ed	feat(backend): add depth-anything (Depth Anything 3) C++/ggml backend + gallery (#10352) ... * feat(backend): add depth-anything (Depth Anything 3) C++/ggml backend + gallery Mirrors the locate-anything-cpp backend to register a new depth-anything backend that wraps the Depth Anything 3 ggml port (depth-anything.cpp) via purego (cgo-less, no Python at inference). - backend/go/depth-anything-cpp/: gRPC backend (Load + Predict + GenerateImage), purego binding to the da_capi_* C ABI, CMake/Makefile/run/package/test scripts building depth-anything.cpp's DA_SHARED static .so per CPU variant. - backend/index.yaml: depth-anything backend meta + all hardware-variant capability entries (cpu/cuda12/cuda13/intel-sycl-f32+f16/vulkan/nvidia-l4t). - gallery/index.yaml: 8 Depth Anything 3 GGUF models (base q4_k/q8_0/f16/f32, small, large, giant, mono-large). - .github/backend-matrix.yml: one build entry per hardware variant. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(depth): typed Depth RPC + REST endpoint exposing full DA3 data Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(depth): pin depth-anything.cpp to e0b6814 (ABI 3 dense C-API) The Depth RPC handler calls da_capi_depth_dense / da_capi_points (C-API ABI 3); pin the native build to the commit that exports them. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(depth): pin depth-anything.cpp to v0.1.0 release (b515c31) Repoint the native version from the now-orphaned e0b6814 to the b515c31 release commit, kept alive by the upstream v0.1.0 tag. C-API is unchanged (da_capi_abi_version == 3). Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(depth): wire depth-anything-cpp into build, CI bump, and importer The backend dir, gallery index, and CI build-matrix were present but the backend was never wired into the integration points that adding-backends.md requires: - root Makefile: add to .NOTPARALLEL, the test-extra chain, a BACKEND_* definition, the docker-build target eval, and docker-build-backends (mirrors parakeet-cpp; the backend's own Makefile already documented that its `test` target is driven by test-extra). - bump_deps.yaml: register the DEPTHANYTHING_VERSION pin so the daily auto-bump bot tracks mudler/depth-anything.cpp master (it cannot see an unregistered Makefile pin). - import form: add a preference-only KnownBackend entry so depth-anything is selectable at /import-model (mirrors sam3-cpp; no reliable GGUF auto-detect signal, so pref-only per the doc's default). changed-backends.js needs no entry: the generic golang suffix branch already resolves backend/go/depth-anything-cpp/. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * feat(depth): auto-detect importer for depth-anything GGUFs Replace the preference-only entry with a real auto-detect importer (mirrors parakeet-cpp / locate-anything): - DepthAnythingImporter matches a .gguf whose name carries a depth-anything token (depth-anything-<size>-<quant>.gguf), so /import-model recognises mudler/depth-anything.cpp-gguf repos and direct GGUF URLs without an explicit backend preference. preferences.backend= "depth-anything" still forces it. - Registered before LlamaCPPImporter so its GGUF bundles aren't claimed by the generic .gguf importer; the narrow name match means it cannot claim arbitrary llama GGUFs or the upstream safetensors PyTorch repos. - Multi-quant repos pick the smallest quant by default (q4_k -> ... -> f32, depth stays >0.998 corr even at q4_k); quantizations preference overrides. - Drops the now-redundant knownPrefOnlyBackends entry (importer-backed backends are not listed there, matching parakeet-cpp). - Table-driven Ginkgo test covers detection, negative cases (llama GGUF, upstream safetensors), default/override/fallback quant pick, and direct URL import. 10/10 specs pass. Assisted-by: Claude:claude-opus-4-8 Signed-off-by: Ettore Di Giacinto <mudler@localai.io> * fix(depth): check conn.Close error in grpc Depth client (errcheck) The new Depth() client method used a bare `defer conn.Close()`. golangci-lint runs with new-from-merge-base, so although the 39 sibling methods use the same bare form (grandfathered), the newly added line trips errcheck. Drop the result explicitly to satisfy the linter. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-4-8 * fix(depth): bump depth-anything.cpp to v0.1.1 (embeddable CMake) v0.1.0 (b515c31) used ${CMAKE_SOURCE_DIR} for its include dirs, which points at the parent project when built via add_subdirectory() as this backend does, so the container build failed with missing stb_image.h / da_gguf_keys.h. v0.1.1 (2d42897) switches to project-relative paths. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-4-8 * fix(depth): resolve gosec findings in the backend wrapper The code-scanning gate flagged three new failure-level alerts in godepthanythingcpp.go (gosec runs with -no-fail; GitHub gates on new alerts): - G301: export dirs were created with 0o755. Tighten to 0o750 (no world access needed for backend-written export output). - G304: writeDepthPNG creates req.GetDst(). That path is chosen by the LocalAI core as the intended output destination (same pattern every image backend uses), not attacker input, so annotate with #nosec G304 and document why. The remaining G103 "audit unsafe" notes on the unsafe.Slice C-buffer copies are warning-level (the same purego interop whisper/parakeet use) and do not gate the check, per the supertonic exclusion precedent in secscan.yaml. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-4-8 * fix(depth): bump depth-anything.cpp to v0.1.2 (CUDA cross-build arch) v0.1.1 forced CMAKE_CUDA_ARCHITECTURES=native, which breaks the GPU-less l4t/cublas CI builds (nvcc "Unsupported gpu architecture 'compute_'" on CMake 3.22). v0.1.2 (442eea4) drops the override and lets ggml pick its default cross-build arch list. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-4-8 --------- Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Co-authored-by: Ettore Di Giacinto <mudler@localai.io>	2026-06-16 16:28:28 +02:00
Richard Palethorpe	085fc53bbc	fix(router): production-ready request router + auto-size batch for embedding/rerank (#10104) ... * fix(router): score classifier production-readiness Conversation trimming runs through the classifier model's chat template and trims by exact token count, sized to the model's n_batch which is now scaled to context so long probes can't crash the backend. Missing chat_message templates are a hard error at router build time. Router- facing factories (Embedder/Scorer/Reranker/TokenCounter) re-resolve ModelConfig per call so a model installed post-startup doesn't bind a stub Backend="" config and silently fall into the loader's auto- iterate path. New 'vector_store' backend trace recorded inside localVectorStore on every Search/Insert — including the backend-load-failure path that previously vanished into an xlog.Warn — with outcome tagging (hit/miss/empty_store/backend_load_error/find_error/insert_error/ok). Companion cleanup drops misleading similarity:0 and input_tokens_count:0 from non-hit and text-mode traces. Gallery local-store-development aliases to 'local-store' so the master image satisfies pkg/model.LocalStoreBackend lookups from the embedding cache. Misc: llama-cpp TokenizeString reads the correct 'prompt' JSON key (the original bug); ModelTokenize nil-guard; non-fatal mitm proxy startup; PII 'route_local' renamed to 'allow' with docs/UI in sync; model-editor footer no longer eats the edit area on small screens; several config-editor template/dropdown/section fixes. Tests: e2e router specs (casual/code-hint + long-conversation trim), vector_store trace specs, lazy-factory specs, gallery dev-alias resolution, Playwright trace badge + scroll regression. Assisted-by: Claude:claude-opus-4-7 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * feat(backend): auto-size batch to context for embedding and rerank models Embedding and rerank models pool over the whole input in a single physical batch (n_ubatch). With batch left at the 512 default, the backend rejects longer inputs with "input is too large to process", silently capping a large-context embedder (e.g. 8k/32k) at 512 tokens. Size n_batch to the context for these single-pass usecases, mirroring the existing FLAG_SCORE behaviour; an explicit batch: still wins. Extracts EffectiveContextSize/EffectiveBatchSize from grpcModelOpts so the effective decode window has one home for other callers to reuse. Adds an e2e-aio regression test that embeds a >512-token input. The AIO embedding model is switched to nomic-embed-text-v1.5 (2048 context) because the previous granite model was capped at 512 tokens and could not exercise the larger batch. Assisted-by: claude-code:claude-opus-4-8 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> * fix(gallery): raise arch-router scoring output cap via parallel:64 Scoring decodes the whole prompt+candidate in a single llama_decode and reads one logit row per candidate token. The vendored llama.cpp server caps causal output rows at n_parallel, so the default of 1 aborts with GGML_ASSERT(n_outputs_max <= cparams.n_outputs_max) on multi-token route labels. Set options: [parallel:64] on both arch-router quant entries to lift the cap; kv_unified (the grpc-server default) keeps the full context per sequence, so this does not split the KV cache. Assisted-by: claude-code:claude-opus-4-8 [Claude Code] Signed-off-by: Richard Palethorpe <io@richiejp.com> --------- Signed-off-by: Richard Palethorpe <io@richiejp.com>	2026-06-12 16:21:15 +02:00