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>

pkg/mcp/localaitools/client.go

@@ -76,24 +76,11 @@ type LocalAIClient interface {
 	GetUsageStats(ctx context.Context, q UsageStatsQuery) (*UsageStats, error)
 
 	// ---- PII filter ----
-	// ListPIIPatterns returns the active PII pattern set with each
-	// one's action.
-	ListPIIPatterns(ctx context.Context) ([]PIIPattern, error)
 	// GetPIIEvents returns recent redaction events. Implementation
-	// enforces "admin required" when auth is on.
+	// enforces "admin required" when auth is on. The regex pattern tools
+	// were removed — detection policy lives on each detector model's
+	// pii_detection block, managed via the model-config tools.
 	GetPIIEvents(ctx context.Context, q PIIEventsQuery) ([]PIIEvent, error)
-	// TestPIIRedaction dry-runs the redactor against text. No event
-	// is recorded.
-	TestPIIRedaction(ctx context.Context, req PIIRedactTestRequest) (*PIIRedactTestResult, error)
-	// SetPIIPatternAction mutates the named pattern's action and/or
-	// disabled state in-process. Transient until PersistPIIPatterns is
-	// called — runtime_settings.json then applies the deltas on the
-	// next start. Admin-required.
-	SetPIIPatternAction(ctx context.Context, req PIIPatternActionUpdate) error
-
-	// PersistPIIPatterns snapshots the live redactor's per-pattern
-	// (action, disabled) state into runtime_settings.json. Admin-required.
-	PersistPIIPatterns(ctx context.Context) error
 
 	// ---- Middleware admin ----
 	// GetMiddlewareStatus returns the aggregated state surfaced on the

pkg/mcp/localaitools/coverage_test.go

@@ -38,25 +38,21 @@ var toolToHTTPRoute = map[string]string{
 	ToolVRAMEstimate:        "POST /api/models/vram-estimate",
 	ToolGetBranding:         "GET /api/branding",
 	ToolGetUsageStats:       "GET /api/usage (or /api/usage/all when all=true)",
-	ToolListPIIPatterns:     "GET /api/pii/patterns",
 	ToolGetPIIEvents:        "GET /api/pii/events",
-	ToolTestPIIRedaction:    "POST /api/pii/test",
 	ToolGetMiddlewareStatus: "GET /api/middleware/status",
 	ToolGetRouterDecisions:  "GET /api/router/decisions",
 
 	// Mutating tools.
-	ToolInstallModel:        "POST /models/apply",
-	ToolImportModelURI:      "POST /models/import-uri",
-	ToolDeleteModel:         "POST /models/delete/:name",
-	ToolEditModelConfig:     "PATCH /api/models/config-json/:name",
-	ToolReloadModels:        "POST /models/reload",
-	ToolInstallBackend:      "POST /backends/apply",
-	ToolUpgradeBackend:      "POST /backends/upgrade/:name",
-	ToolToggleModelState:    "PUT /models/toggle-state/:name/:action",
-	ToolToggleModelPinned:   "PUT /models/toggle-pinned/:name/:action",
-	ToolSetBranding:         "POST /api/settings (instance_name, instance_tagline)",
-	ToolSetPIIPatternAction: "PUT /api/pii/patterns/:id",
-	ToolPersistPIIPatterns:  "POST /api/pii/patterns/persist",
+	ToolInstallModel:      "POST /models/apply",
+	ToolImportModelURI:    "POST /models/import-uri",
+	ToolDeleteModel:       "POST /models/delete/:name",
+	ToolEditModelConfig:   "PATCH /api/models/config-json/:name",
+	ToolReloadModels:      "POST /models/reload",
+	ToolInstallBackend:    "POST /backends/apply",
+	ToolUpgradeBackend:    "POST /backends/upgrade/:name",
+	ToolToggleModelState:  "PUT /models/toggle-state/:name/:action",
+	ToolToggleModelPinned: "PUT /models/toggle-pinned/:name/:action",
+	ToolSetBranding:       "POST /api/settings (instance_name, instance_tagline)",
 }
 
 // allKnownTools is the union of expectedFullCatalog (defined in

pkg/mcp/localaitools/dto.go

@@ -77,11 +77,11 @@ type Backend struct {
 
 // SystemInfo summarises the LocalAI deployment.
 type SystemInfo struct {
-	Version          string   `json:"version"`
-	Distributed      bool     `json:"distributed"`
-	BackendsPath     string   `json:"backends_path,omitempty"`
-	ModelsPath       string   `json:"models_path,omitempty"`
-	LoadedModels     []string `json:"loaded_models,omitempty"`
+	Version           string   `json:"version"`
+	Distributed       bool     `json:"distributed"`
+	BackendsPath      string   `json:"backends_path,omitempty"`
+	ModelsPath        string   `json:"models_path,omitempty"`
+	LoadedModels      []string `json:"loaded_models,omitempty"`
 	InstalledBackends []string `json:"installed_backends,omitempty"`
 }
 
@@ -184,19 +184,11 @@ type UsageBucket struct {
 
 // ---- PII / sensitive data tools ----
 
-// PIIPattern is one row in the list_pii_patterns response.
-type PIIPattern struct {
-	ID             string `json:"id"`
-	Description    string `json:"description"`
-	Action         string `json:"action"` // mask | block | allow
-	MaxMatchLength int    `json:"max_match_length"`
-}
-
 // PIIEventsQuery filters get_pii_events.
 type PIIEventsQuery struct {
 	CorrelationID string `json:"correlation_id,omitempty" jsonschema:"Optional X-Correlation-ID join key (binds events to the request and usage record)."`
 	UserID        string `json:"user_id,omitempty"        jsonschema:"Optional user id to scope the query."`
-	PatternID     string `json:"pattern_id,omitempty"     jsonschema:"Optional pattern id (e.g. email, ssn)."`
+	PatternID     string `json:"pattern_id,omitempty"     jsonschema:"Optional detector group id (e.g. ner:EMAIL)."`
 	Limit         int    `json:"limit,omitempty"          jsonschema:"Maximum events. Defaults to 100."`
 }
 
@@ -215,38 +207,6 @@ type PIIEvent struct {
 	CreatedAt     string `json:"created_at"`
 }
 
-// PIIRedactTestRequest is the input for test_pii_redaction.
-type PIIRedactTestRequest struct {
-	Text string `json:"text" jsonschema:"The candidate text. Will be run through the redactor without recording an event."`
-}
-
-// PIIRedactTestResult is the output for test_pii_redaction. spans
-// describes where the redactor matched; redacted is the text after
-// applying mask actions; blocked / masked flag what was done.
-type PIIRedactTestResult struct {
-	Redacted string         `json:"redacted"`
-	Spans    []PIIEventSpan `json:"spans"`
-	Blocked  bool           `json:"blocked"`
-	Masked   bool           `json:"masked"`
-}
-
-type PIIEventSpan struct {
-	Start      int    `json:"start"`
-	End        int    `json:"end"`
-	Pattern    string `json:"pattern"`
-	HashPrefix string `json:"hash_prefix"`
-}
-
-// PIIPatternActionUpdate is the input for set_pii_pattern_action.
-// At least one of Action or Disabled must be set. Mutations are
-// transient by default — call persist_pii_patterns to flush them
-// to runtime_settings.json so the next start re-applies them.
-type PIIPatternActionUpdate struct {
-	ID       string `json:"id" jsonschema:"Pattern id to mutate (e.g. email, ssn, credit_card, api_key_prefix)."`
-	Action   string `json:"action,omitempty" jsonschema:"New action: mask, block, or allow. Optional — omit to leave the action unchanged."`
-	Disabled *bool  `json:"disabled,omitempty" jsonschema:"Set true to skip this pattern entirely; false to re-enable. Optional — omit to leave enabled-state unchanged."`
-}
-
 // MiddlewareStatus is the aggregated /api/middleware/status payload —
 // the React Middleware page renders this in one go. Routing is a
 // placeholder until subsystem 2 lands.
@@ -255,25 +215,25 @@ type MiddlewareStatus struct {
 	Router MiddlewareRouterStatus `json:"router"`
 }
 
-// MiddlewarePIIStatus shows what the redactor is doing right now and
-// which models opt in. enabled_globally=false means --disable-pii.
+// MiddlewarePIIStatus shows which models opt in to PII redaction and the
+// NER detector models they reference. The detection policy itself lives
+// on each detector model's pii_detection block.
 type MiddlewarePIIStatus struct {
-	EnabledGlobally           bool                  `json:"enabled_globally"`
-	Reason                    string                `json:"reason,omitempty"`
-	DefaultEnabledForBackends []string              `json:"default_enabled_for_backends,omitempty"`
-	Patterns                  []PIIPattern          `json:"patterns"`
-	Models                    []MiddlewarePIIModel  `json:"models"`
-	RecentEventCount          int                   `json:"recent_event_count"`
+	EnabledGlobally           bool                 `json:"enabled_globally"`
+	Reason                    string               `json:"reason,omitempty"`
+	DefaultEnabledForBackends []string             `json:"default_enabled_for_backends,omitempty"`
+	Models                    []MiddlewarePIIModel `json:"models"`
+	RecentEventCount          int                  `json:"recent_event_count"`
 }
 
 // MiddlewarePIIModel is one model row in the per-model PII table.
 type MiddlewarePIIModel struct {
-	Name              string            `json:"name"`
-	Backend           string            `json:"backend"`
-	Enabled           bool              `json:"enabled"`
-	Explicit          bool              `json:"explicit"`             // Did YAML set Enabled, or did the backend prefix decide?
-	DefaultForBackend bool              `json:"default_for_backend"`  // Backend matches the auto-on rule (proxy-*).
-	Overrides         map[string]string `json:"overrides,omitempty"`
+	Name              string   `json:"name"`
+	Backend           string   `json:"backend"`
+	Enabled           bool     `json:"enabled"`
+	Explicit          bool     `json:"explicit"`            // Did YAML set Enabled, or did the backend prefix decide?
+	DefaultForBackend bool     `json:"default_for_backend"` // Backend matches the auto-on rule (proxy-*).
+	Detectors         []string `json:"detectors,omitempty"` // NER detector model names this config references.
 }
 
 // MiddlewareRouterStatus is the placeholder shape the Routing tab

pkg/mcp/localaitools/fakes_test.go

@@ -45,10 +45,7 @@ type fakeClient struct {
 	getBranding         func() (*Branding, error)
 	setBranding         func(SetBrandingRequest) (*Branding, error)
 	getUsageStats       func(UsageStatsQuery) (*UsageStats, error)
-	listPIIPatterns     func() ([]PIIPattern, error)
 	getPIIEvents        func(PIIEventsQuery) ([]PIIEvent, error)
-	testPIIRedaction    func(PIIRedactTestRequest) (*PIIRedactTestResult, error)
-	setPIIPatternAction func(PIIPatternActionUpdate) error
 	getMiddlewareStatus func() (*MiddlewareStatus, error)
 	getRouterDecisions  func(RouterDecisionsQuery) ([]RouterDecision, error)
 }
@@ -253,14 +250,6 @@ func (f *fakeClient) GetUsageStats(_ context.Context, q UsageStatsQuery) (*Usage
 	}, nil
 }
 
-func (f *fakeClient) ListPIIPatterns(_ context.Context) ([]PIIPattern, error) {
-	f.record("ListPIIPatterns", nil)
-	if f.listPIIPatterns != nil {
-		return f.listPIIPatterns()
-	}
-	return []PIIPattern{}, nil
-}
-
 func (f *fakeClient) GetPIIEvents(_ context.Context, q PIIEventsQuery) ([]PIIEvent, error) {
 	f.record("GetPIIEvents", q)
 	if f.getPIIEvents != nil {
@@ -269,27 +258,6 @@ func (f *fakeClient) GetPIIEvents(_ context.Context, q PIIEventsQuery) ([]PIIEve
 	return []PIIEvent{}, nil
 }
 
-func (f *fakeClient) TestPIIRedaction(_ context.Context, req PIIRedactTestRequest) (*PIIRedactTestResult, error) {
-	f.record("TestPIIRedaction", req)
-	if f.testPIIRedaction != nil {
-		return f.testPIIRedaction(req)
-	}
-	return &PIIRedactTestResult{Redacted: req.Text}, nil
-}
-
-func (f *fakeClient) SetPIIPatternAction(_ context.Context, req PIIPatternActionUpdate) error {
-	f.record("SetPIIPatternAction", req)
-	if f.setPIIPatternAction != nil {
-		return f.setPIIPatternAction(req)
-	}
-	return nil
-}
-
-func (f *fakeClient) PersistPIIPatterns(_ context.Context) error {
-	f.record("PersistPIIPatterns", nil)
-	return nil
-}
-
 func (f *fakeClient) GetRouterDecisions(_ context.Context, q RouterDecisionsQuery) ([]RouterDecision, error) {
 	f.record("GetRouterDecisions", q)
 	if f.getRouterDecisions != nil {
@@ -306,10 +274,8 @@ func (f *fakeClient) GetMiddlewareStatus(_ context.Context) (*MiddlewareStatus,
 	return &MiddlewareStatus{
 		PII: MiddlewarePIIStatus{
 			EnabledGlobally: true,
-			Patterns:        []PIIPattern{},
 			Models:          []MiddlewarePIIModel{},
 		},
 		Router: MiddlewareRouterStatus{Configured: false, Models: []string{}},
 	}, nil
 }
-

pkg/mcp/localaitools/httpapi/client.go

@@ -582,16 +582,6 @@ func (c *Client) GetUsageStats(ctx context.Context, q localaitools.UsageStatsQue
 
 // ---- PII filter ----
 
-func (c *Client) ListPIIPatterns(ctx context.Context) ([]localaitools.PIIPattern, error) {
-	var raw struct {
-		Patterns []localaitools.PIIPattern `json:"patterns"`
-	}
-	if err := c.do(ctx, http.MethodGet, routePIIPatterns, nil, &raw); err != nil {
-		return nil, err
-	}
-	return raw.Patterns, nil
-}
-
 func (c *Client) GetPIIEvents(ctx context.Context, q localaitools.PIIEventsQuery) ([]localaitools.PIIEvent, error) {
 	qs := url.Values{}
 	if q.CorrelationID != "" {
@@ -624,35 +614,6 @@ func (c *Client) GetPIIEvents(ctx context.Context, q localaitools.PIIEventsQuery
 	return raw.Events, nil
 }
 
-func (c *Client) TestPIIRedaction(ctx context.Context, req localaitools.PIIRedactTestRequest) (*localaitools.PIIRedactTestResult, error) {
-	var out localaitools.PIIRedactTestResult
-	if err := c.do(ctx, http.MethodPost, routePIITest, map[string]string{"text": req.Text}, &out); err != nil {
-		return nil, err
-	}
-	return &out, nil
-}
-
-func (c *Client) SetPIIPatternAction(ctx context.Context, req localaitools.PIIPatternActionUpdate) error {
-	if req.ID == "" {
-		return fmt.Errorf("pattern id is required")
-	}
-	body := map[string]any{}
-	if req.Action != "" {
-		body["action"] = req.Action
-	}
-	if req.Disabled != nil {
-		body["disabled"] = *req.Disabled
-	}
-	if len(body) == 0 {
-		return fmt.Errorf("must specify action and/or disabled")
-	}
-	return c.do(ctx, http.MethodPut, routePIIPatternByID(req.ID), body, nil)
-}
-
-func (c *Client) PersistPIIPatterns(ctx context.Context) error {
-	return c.do(ctx, http.MethodPost, routePIIPatternsPersist, nil, nil)
-}
-
 func (c *Client) GetMiddlewareStatus(ctx context.Context) (*localaitools.MiddlewareStatus, error) {
 	var out localaitools.MiddlewareStatus
 	if err := c.do(ctx, http.MethodGet, routeMiddleware, nil, &out); err != nil {

pkg/mcp/localaitools/httpapi/routes.go

@@ -11,33 +11,26 @@ import (
 // registrations in core/http/routes/localai.go — the Tool↔REST drift detector
 // in coverage_test.go documents the mapping.
 const (
-	routeWelcome            = "/"
-	routeModelsApply        = "/models/apply"
-	routeModelsAvail        = "/models/available"
-	routeModelsGall         = "/models/galleries"
-	routeModelsImport       = "/models/import-uri"
-	routeModelsReload       = "/models/reload"
-	routeBackends           = "/backends"
-	routeBackendsKnown      = "/backends/known"
-	routeBackendsApply      = "/backends/apply"
-	routeNodes              = "/api/nodes"
-	routeVRAMEstimate       = "/api/models/vram-estimate"
-	routeBranding           = "/api/branding"
-	routeSettings           = "/api/settings"
-	routeUsage              = "/api/usage"
-	routeUsageAll           = "/api/usage/all"
-	routePIIPatterns        = "/api/pii/patterns"
-	routePIIPatternsPersist = "/api/pii/patterns/persist"
-	routePIIEvents          = "/api/pii/events"
-	routePIITest            = "/api/pii/test"
-	routeMiddleware         = "/api/middleware/status"
-	routeRouterDecisions    = "/api/router/decisions"
+	routeWelcome         = "/"
+	routeModelsApply     = "/models/apply"
+	routeModelsAvail     = "/models/available"
+	routeModelsGall      = "/models/galleries"
+	routeModelsImport    = "/models/import-uri"
+	routeModelsReload    = "/models/reload"
+	routeBackends        = "/backends"
+	routeBackendsKnown   = "/backends/known"
+	routeBackendsApply   = "/backends/apply"
+	routeNodes           = "/api/nodes"
+	routeVRAMEstimate    = "/api/models/vram-estimate"
+	routeBranding        = "/api/branding"
+	routeSettings        = "/api/settings"
+	routeUsage           = "/api/usage"
+	routeUsageAll        = "/api/usage/all"
+	routePIIEvents       = "/api/pii/events"
+	routeMiddleware      = "/api/middleware/status"
+	routeRouterDecisions = "/api/router/decisions"
 )
 
-func routePIIPatternByID(id string) string {
-	return "/api/pii/patterns/" + url.PathEscape(id)
-}
-
 func routeJobStatus(jobID string) string {
 	return "/models/jobs/" + url.PathEscape(jobID)
 }

pkg/mcp/localaitools/inproc/client.go

@@ -14,10 +14,10 @@ import (
 	"github.com/mudler/LocalAI/core/config"
 	"github.com/mudler/LocalAI/core/gallery"
 	"github.com/mudler/LocalAI/core/gallery/importers"
+	"github.com/mudler/LocalAI/core/http/auth"
 	"github.com/mudler/LocalAI/core/schema"
 	"github.com/mudler/LocalAI/core/services/galleryop"
 	"github.com/mudler/LocalAI/core/services/modeladmin"
-	"github.com/mudler/LocalAI/core/http/auth"
 	"github.com/mudler/LocalAI/core/services/routing/billing"
 	"github.com/mudler/LocalAI/core/services/routing/pii"
 	"github.com/mudler/LocalAI/core/services/routing/router"
@@ -619,23 +619,6 @@ func (c *Client) GetUsageStats(ctx context.Context, q localaitools.UsageStatsQue
 
 // ---- PII filter ----
 
-func (c *Client) ListPIIPatterns(_ context.Context) ([]localaitools.PIIPattern, error) {
-	if c.PIIRedactor == nil {
-		return nil, errors.New("PII filter is disabled")
-	}
-	patterns := c.PIIRedactor.Patterns()
-	out := make([]localaitools.PIIPattern, 0, len(patterns))
-	for _, p := range patterns {
-		out = append(out, localaitools.PIIPattern{
-			ID:             p.ID,
-			Description:    p.Description,
-			Action:         string(p.Action),
-			MaxMatchLength: p.MaxMatchLength,
-		})
-	}
-	return out, nil
-}
-
 func (c *Client) GetPIIEvents(ctx context.Context, q localaitools.PIIEventsQuery) ([]localaitools.PIIEvent, error) {
 	if c.PIIEvents == nil {
 		return nil, errors.New("PII filter is disabled")
@@ -668,77 +651,6 @@ func (c *Client) GetPIIEvents(ctx context.Context, q localaitools.PIIEventsQuery
 	return out, nil
 }
 
-func (c *Client) SetPIIPatternAction(_ context.Context, req localaitools.PIIPatternActionUpdate) error {
-	if c.PIIRedactor == nil {
-		return errors.New("PII filter is disabled")
-	}
-	if req.ID == "" {
-		return errors.New("pattern id is required")
-	}
-	if req.Action == "" && req.Disabled == nil {
-		return errors.New("must specify action and/or disabled")
-	}
-	if req.Action != "" {
-		if err := c.PIIRedactor.SetAction(req.ID, pii.Action(req.Action)); err != nil {
-			return err
-		}
-	}
-	if req.Disabled != nil {
-		if err := c.PIIRedactor.SetDisabled(req.ID, *req.Disabled); err != nil {
-			return err
-		}
-	}
-	return nil
-}
-
-// PersistPIIPatterns snapshots the current redactor state into
-// runtime_settings.json. Mirrors POST /api/pii/patterns/persist.
-func (c *Client) PersistPIIPatterns(_ context.Context) error {
-	if c.PIIRedactor == nil {
-		return errors.New("PII filter is disabled")
-	}
-	if c.AppConfig == nil {
-		return errors.New("app config not available")
-	}
-	existing, err := c.AppConfig.ReadPersistedSettings()
-	if err != nil {
-		return fmt.Errorf("read settings: %w", err)
-	}
-	defaults, err := pii.LoadConfig(c.AppConfig.PIIConfigPath)
-	if err != nil {
-		return fmt.Errorf("reload defaults: %w", err)
-	}
-	defaultByID := make(map[string]pii.Pattern, len(defaults))
-	for _, d := range defaults {
-		defaultByID[d.ID] = d
-	}
-	overrides := map[string]config.PIIPatternRuntimeOverride{}
-	for _, p := range c.PIIRedactor.Patterns() {
-		d, known := defaultByID[p.ID]
-		ov := config.PIIPatternRuntimeOverride{}
-		changed := false
-		if !known || p.Action != d.Action {
-			action := string(p.Action)
-			ov.Action = &action
-			changed = true
-		}
-		if !known || p.Disabled != d.Disabled {
-			disabled := p.Disabled
-			ov.Disabled = &disabled
-			changed = true
-		}
-		if changed {
-			overrides[p.ID] = ov
-		}
-	}
-	existing.PIIPatternOverrides = &overrides
-	if err := c.AppConfig.WritePersistedSettings(existing); err != nil {
-		return fmt.Errorf("write settings: %w", err)
-	}
-	c.AppConfig.PIIPatternOverrides = overrides
-	return nil
-}
-
 func (c *Client) GetRouterDecisions(ctx context.Context, q localaitools.RouterDecisionsQuery) ([]localaitools.RouterDecision, error) {
 	if c.RouterDecisions == nil {
 		return []localaitools.RouterDecision{}, nil
@@ -779,23 +691,10 @@ func (c *Client) GetMiddlewareStatus(ctx context.Context) (*localaitools.Middlew
 		Note:       "Intelligent routing is not yet implemented.",
 	}
 	piiSection := localaitools.MiddlewarePIIStatus{
-		EnabledGlobally: c.PIIRedactor != nil,
-		Patterns:        []localaitools.PIIPattern{},
+		EnabledGlobally: c.PIIEvents != nil,
 		Models:          []localaitools.MiddlewarePIIModel{},
 	}
-	if c.PIIRedactor == nil {
-		piiSection.Reason = "--disable-pii"
-		return &localaitools.MiddlewareStatus{PII: piiSection, Router: router}, nil
-	}
 	piiSection.DefaultEnabledForBackends = []string{"cloud-proxy"}
-	for _, p := range c.PIIRedactor.Patterns() {
-		piiSection.Patterns = append(piiSection.Patterns, localaitools.PIIPattern{
-			ID:             p.ID,
-			Description:    p.Description,
-			Action:         string(p.Action),
-			MaxMatchLength: p.MaxMatchLength,
-		})
-	}
 	if c.ConfigLoader != nil {
 		for _, cfg := range c.ConfigLoader.GetAllModelsConfigs() {
 			cfg := cfg
@@ -805,7 +704,7 @@ func (c *Client) GetMiddlewareStatus(ctx context.Context) (*localaitools.Middlew
 				Enabled:           cfg.PIIIsEnabled(),
 				Explicit:          cfg.PII.Enabled != nil,
 				DefaultForBackend: cfg.Backend == "cloud-proxy",
-				Overrides:         cfg.PIIPatternOverrides(),
+				Detectors:         cfg.PIIDetectors(),
 			})
 		}
 	}
@@ -817,27 +716,6 @@ func (c *Client) GetMiddlewareStatus(ctx context.Context) (*localaitools.Middlew
 	return &localaitools.MiddlewareStatus{PII: piiSection, Router: router}, nil
 }
 
-func (c *Client) TestPIIRedaction(_ context.Context, req localaitools.PIIRedactTestRequest) (*localaitools.PIIRedactTestResult, error) {
-	if c.PIIRedactor == nil {
-		return nil, errors.New("PII filter is disabled")
-	}
-	res := c.PIIRedactor.Redact(req.Text)
-	out := &localaitools.PIIRedactTestResult{
-		Redacted: res.Redacted,
-		Blocked:  res.Blocked,
-		Masked:   res.Masked,
-	}
-	for _, s := range res.Spans {
-		out.Spans = append(out.Spans, localaitools.PIIEventSpan{
-			Start:      s.Start,
-			End:        s.End,
-			Pattern:    s.Pattern,
-			HashPrefix: s.HashPrefix,
-		})
-	}
-	return out, nil
-}
-
 func capabilityFlagsOf(m *config.ModelConfig) []string {
 	var out []string
 	for label, flag := range config.GetAllModelConfigUsecases() {

pkg/mcp/localaitools/server_test.go

@@ -91,13 +91,9 @@ var expectedFullCatalog = sortedStrings(
 	ToolListInstalledModels,
 	ToolListKnownBackends,
 	ToolListNodes,
-	ToolListPIIPatterns,
-	ToolPersistPIIPatterns,
 	ToolReloadModels,
 	ToolSetBranding,
-	ToolSetPIIPatternAction,
 	ToolSystemInfo,
-	ToolTestPIIRedaction,
 	ToolToggleModelPinned,
 	ToolToggleModelState,
 	ToolUpgradeBackend,
@@ -119,9 +115,7 @@ var expectedReadOnlyCatalog = sortedStrings(
 	ToolListInstalledModels,
 	ToolListKnownBackends,
 	ToolListNodes,
-	ToolListPIIPatterns,
 	ToolSystemInfo,
-	ToolTestPIIRedaction,
 	ToolVRAMEstimate,
 )
 

pkg/mcp/localaitools/tools.go

@@ -20,9 +20,7 @@ const (
 	ToolVRAMEstimate        = "vram_estimate"
 	ToolGetBranding         = "get_branding"
 	ToolGetUsageStats       = "get_usage_stats"
-	ToolListPIIPatterns     = "list_pii_patterns"
 	ToolGetPIIEvents        = "get_pii_events"
-	ToolTestPIIRedaction    = "test_pii_redaction"
 	ToolGetMiddlewareStatus = "get_middleware_status"
 	ToolGetRouterDecisions  = "get_router_decisions"
 
@@ -38,8 +36,6 @@ const (
 	ToolToggleModelState  = "toggle_model_state"
 	ToolToggleModelPinned = "toggle_model_pinned"
 	ToolSetBranding       = "set_branding"
-	ToolSetPIIPatternAction = "set_pii_pattern_action"
-	ToolPersistPIIPatterns  = "persist_pii_patterns"
 )
 
 // DefaultServerName is the MCP Implementation.Name surfaced when

pkg/mcp/localaitools/tools_middleware.go

@@ -7,21 +7,21 @@ import (
 )
 
 // registerMiddlewareTools wires the routing-module admin surface for the
-// MCP server. The two tools mirror what the React /app/middleware page
-// exposes:
+// MCP server, mirroring what the React /app/middleware page exposes:
 //
 //   - get_middleware_status: read-only aggregator. The agent can ask
-//     "what's filtering my requests?" and get back the active PII
-//     pattern set, the per-model resolved enabled/override state, and
-//     a placeholder for routing.
-//   - set_pii_pattern_action: mutating. Mutations are TRANSIENT — they
-//     live until process restart, when patterns reload from the YAML
-//     defaults. The skill prompt should warn the user about that
-//     before applying lasting changes.
-func registerMiddlewareTools(s *mcp.Server, client LocalAIClient, opts Options) {
+//     "what's filtering my requests?" and get back the per-model PII
+//     enabled state + the detector models each references, recent event
+//     count, plus the active router models and their classifier configs.
+//   - get_router_decisions: read-only routing-decision log.
+//
+// PII detection policy lives on each detector model's pii_detection
+// block, edited via the model-config tools — there is no global pattern
+// set to mutate here anymore.
+func registerMiddlewareTools(s *mcp.Server, client LocalAIClient, _ Options) {
 	mcp.AddTool(s, &mcp.Tool{
 		Name:        ToolGetMiddlewareStatus,
-		Description: "Aggregated routing-module status: PII pattern catalogue with current actions, per-model resolved PII state and overrides, recent event count, plus the active router models and their classifier configs. Read-only.",
+		Description: "Aggregated routing-module status: per-model resolved PII state and the NER detector models each one references, recent event count, plus the active router models and their classifier configs. Read-only.",
 	}, func(ctx context.Context, _ *mcp.CallToolRequest, _ struct{}) (*mcp.CallToolResult, any, error) {
 		status, err := client.GetMiddlewareStatus(ctx)
 		if err != nil {
@@ -40,39 +40,4 @@ func registerMiddlewareTools(s *mcp.Server, client LocalAIClient, opts Options)
 		}
 		return jsonResult(decisions), nil, nil
 	})
-
-	if opts.DisableMutating {
-		return
-	}
-
-	mcp.AddTool(s, &mcp.Tool{
-		Name:        ToolSetPIIPatternAction,
-		Description: "Change a PII pattern's action (mask|block|allow) and/or disabled state in-process. TRANSIENT: the mutation is lost on restart unless followed by persist_pii_patterns. Admin-required.",
-	}, func(ctx context.Context, _ *mcp.CallToolRequest, args PIIPatternActionUpdate) (*mcp.CallToolResult, any, error) {
-		if args.ID == "" {
-			return errorResultf("id is required"), nil, nil
-		}
-		if args.Action == "" && args.Disabled == nil {
-			return errorResultf("at least one of action (mask, block, allow) or disabled must be set"), nil, nil
-		}
-		if err := client.SetPIIPatternAction(ctx, args); err != nil {
-			return errorResult(err), nil, nil
-		}
-		return jsonResult(map[string]any{
-			"id":        args.ID,
-			"action":    args.Action,
-			"disabled":  args.Disabled,
-			"persisted": false,
-		}), nil, nil
-	})
-
-	mcp.AddTool(s, &mcp.Tool{
-		Name:        ToolPersistPIIPatterns,
-		Description: "Snapshot the live PII redactor's per-pattern (action, disabled) state into runtime_settings.json so it re-applies on the next process start. Pairs with set_pii_pattern_action — that one is in-process; this one persists. Admin-required.",
-	}, func(ctx context.Context, _ *mcp.CallToolRequest, _ struct{}) (*mcp.CallToolResult, any, error) {
-		if err := client.PersistPIIPatterns(ctx); err != nil {
-			return errorResult(err), nil, nil
-		}
-		return jsonResult(map[string]any{"persisted": true}), nil, nil
-	})
 }

pkg/mcp/localaitools/tools_pii.go

@@ -7,20 +7,13 @@ import (
 )
 
 func registerPIITools(s *mcp.Server, client LocalAIClient, _ Options) {
-	mcp.AddTool(s, &mcp.Tool{
-		Name:        ToolListPIIPatterns,
-		Description: "List the active PII regex pattern set. Each entry shows the pattern id, description, and current action (mask, block, allow). Read-only.",
-	}, func(ctx context.Context, _ *mcp.CallToolRequest, _ struct{}) (*mcp.CallToolResult, any, error) {
-		patterns, err := client.ListPIIPatterns(ctx)
-		if err != nil {
-			return errorResult(err), nil, nil
-		}
-		return jsonResult(patterns), nil, nil
-	})
-
+	// The regex pattern tools (list/test/set/persist) were removed with
+	// the regex tier. Detection policy now lives on each detector model's
+	// pii_detection block (managed via the model config tools/UI), so the
+	// only PII tool is the read-only audit-event view.
 	mcp.AddTool(s, &mcp.Tool{
 		Name:        ToolGetPIIEvents,
-		Description: "Recent PII redaction events. Filter by correlation_id (joins to a usage record), user_id, or pattern_id. Events never carry the matched value — only an 8-char sha256 prefix so admins can dedupe recurring leaks.",
+		Description: "Recent PII redaction events. Filter by correlation_id (joins to a usage record), user_id, or pattern_id (e.g. ner:EMAIL). Events never carry the matched value — only an 8-char sha256 prefix so admins can dedupe recurring leaks.",
 	}, func(ctx context.Context, _ *mcp.CallToolRequest, args PIIEventsQuery) (*mcp.CallToolResult, any, error) {
 		events, err := client.GetPIIEvents(ctx, args)
 		if err != nil {
@@ -28,18 +21,4 @@ func registerPIITools(s *mcp.Server, client LocalAIClient, _ Options) {
 		}
 		return jsonResult(events), nil, nil
 	})
-
-	mcp.AddTool(s, &mcp.Tool{
-		Name:        ToolTestPIIRedaction,
-		Description: "Dry-run the PII redactor against text without recording a real event. Useful for tuning patterns: paste a candidate string and see whether it would be masked, blocked, or routed locally.",
-	}, func(ctx context.Context, _ *mcp.CallToolRequest, args PIIRedactTestRequest) (*mcp.CallToolResult, any, error) {
-		if args.Text == "" {
-			return errorResultf("text is required"), nil, nil
-		}
-		res, err := client.TestPIIRedaction(ctx, args)
-		if err != nil {
-			return errorResult(err), nil, nil
-		}
-		return jsonResult(res), nil, nil
-	})
 }