server: add tests and fix isHuggingFaceURL edge case

- Add comprehensive tests for isHuggingFaceURL and getNumDownloadParts - Fix bug where domains ending in huggingface.co (like nothuggingface.co) would incorrectly match as HuggingFace URLs - Improve code comments with more detailed documentation
server: reduce download concurrency for HuggingFace URLs
2026-01-19 21:08:16 -05:00 · 2026-01-18 16:45:17 -08:00 · 2026-01-18 16:38:49 -08:00 · 2026-01-13 11:25:31 -08:00 · 2026-01-13 09:13:09 -08:00 · 2026-01-12 22:38:10 -08:00
17 changed files with 406 additions and 2255 deletions
--- a/.github/workflows/release.yaml
+++ b/.github/workflows/release.yaml
@@ -372,13 +372,17 @@ jobs:
          outputs: type=local,dest=dist/${{ matrix.os }}-${{ matrix.arch }}
          cache-from: type=registry,ref=${{ vars.DOCKER_REPO }}:latest
          cache-to: type=inline
      - name: Deduplicate CUDA libraries
        run: |
          ./scripts/deduplicate_cuda_libs.sh dist/${{ matrix.os }}-${{ matrix.arch }}
      - run: |
          for COMPONENT in bin/* lib/ollama/*; do
            case "$COMPONENT" in
-              bin/ollama)                echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
+              bin/ollama*)               echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
              lib/ollama/*.so*)          echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
              lib/ollama/cuda_v*)        echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
              lib/ollama/vulkan*)        echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
              lib/ollama/mlx*)           echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}.tar.in ;;
              lib/ollama/cuda_jetpack5)  echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-jetpack5.tar.in ;;
              lib/ollama/cuda_jetpack6)  echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-jetpack6.tar.in ;;
              lib/ollama/rocm)           echo $COMPONENT >>ollama-${{ matrix.os }}-${{ matrix.arch }}-rocm.tar.in ;;
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -48,9 +48,10 @@ if((CMAKE_OSX_ARCHITECTURES AND NOT CMAKE_OSX_ARCHITECTURES MATCHES "arm64")
    set(GGML_CPU_ALL_VARIANTS ON)
 endif()
-if (CMAKE_OSX_ARCHITECTURES MATCHES "x86_64")
+if(APPLE)
    set(CMAKE_BUILD_RPATH "@loader_path")
    set(CMAKE_INSTALL_RPATH "@loader_path")
    set(CMAKE_BUILD_WITH_INSTALL_RPATH ON)
 endif()
 set(OLLAMA_BUILD_DIR ${CMAKE_BINARY_DIR}/lib/ollama)
@@ -196,6 +197,14 @@ if(MLX_ENGINE)
        FRAMEWORK DESTINATION ${OLLAMA_INSTALL_DIR} COMPONENT MLX
    )
    # Install the Metal library for macOS arm64 (must be colocated with the binary)
    # Metal backend is only built for arm64, not x86_64
    if(APPLE AND CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64")
        install(FILES ${CMAKE_BINARY_DIR}/_deps/mlx-build/mlx/backend/metal/kernels/mlx.metallib
            DESTINATION ${OLLAMA_INSTALL_DIR}
            COMPONENT MLX)
    endif()
    # Manually install cudart and cublas since they might not be picked up as direct dependencies
    if(CUDAToolkit_FOUND)
        file(GLOB CUDART_LIBS
--- a/4
+++ b/4
@@ -161,6 +161,9 @@ ARG GOFLAGS="'-ldflags=-w -s'"
 ENV CGO_ENABLED=1
 ARG CGO_CFLAGS
 ARG CGO_CXXFLAGS
 RUN mkdir -p dist/bin
 RUN --mount=type=cache,target=/root/.cache/go-build \
    go build -tags mlx -trimpath -buildmode=pie -o dist/bin/ollama-mlx .
 FROM base AS build
 WORKDIR /go/src/github.com/ollama/ollama
@@ -182,6 +185,7 @@ COPY --from=cuda-12 dist/lib/ollama /lib/ollama/
 COPY --from=cuda-13 dist/lib/ollama /lib/ollama/
 COPY --from=vulkan  dist/lib/ollama  /lib/ollama/
 COPY --from=mlx     /go/src/github.com/ollama/ollama/dist/lib/ollama /lib/ollama/
 COPY --from=mlx     /go/src/github.com/ollama/ollama/dist/bin/ /bin/
 FROM --platform=linux/arm64 scratch AS arm64
 # COPY --from=cuda-11 dist/lib/ollama/ /lib/ollama/
--- a/cmd/apps.go
+++ b/cmd/apps.go
@@ -1,800 +0,0 @@
 package cmd
 import (
 	"bytes"
 	"context"
 	"encoding/json"
 	"errors"
 	"fmt"
 	"os"
 	"os/exec"
 	"path/filepath"
 	"slices"
 	"strings"
 	"github.com/ollama/ollama/api"
 	"github.com/spf13/cobra"
 )
 type EnvVar struct {
 	Name  string
 	Value string
 }
 type AppConfig struct {
 	Name         string
 	DisplayName  string
 	Command      string
 	EnvVars      func(model string) []EnvVar
 	Args         func(model string) []string
 	Setup        func(models []string) error
 	CheckInstall func() error
 }
 // checkCommand returns an error if the command is not installed
 func checkCommand(cmd, installInstructions string) func() error {
 	return func() error {
 		if _, err := exec.LookPath(cmd); err != nil {
 			return fmt.Errorf("%s is not installed. %s", cmd, installInstructions)
 		}
 		return nil
 	}
 }
 var ClaudeConfig = &AppConfig{
 	Name:        "Claude",
 	DisplayName: "Claude Code",
 	Command:     "claude",
 	EnvVars: func(model string) []EnvVar {
 		return []EnvVar{
 			{Name: "ANTHROPIC_BASE_URL", Value: "http://localhost:11434"},
 			{Name: "ANTHROPIC_API_KEY", Value: "ollama"},
 			{Name: "ANTHROPIC_AUTH_TOKEN", Value: "ollama"},
 		}
 	},
 	Args: func(model string) []string {
 		if model == "" {
 			return nil
 		}
 		return []string{"--model", model}
 	},
 	CheckInstall: checkCommand("claude", "Install with: npm install -g @anthropic-ai/claude-code"),
 }
 var CodexConfig = &AppConfig{
 	Name:        "Codex",
 	DisplayName: "Codex",
 	Command:     "codex",
 	EnvVars: func(model string) []EnvVar {
 		return []EnvVar{}
 	},
 	Args: func(model string) []string {
 		// Defaults to gpt-oss:20b in codex
 		if model == "" {
 			return []string{"--oss"}
 		}
 		return []string{"--oss", "-m", model}
 	},
 	CheckInstall: checkCommand("codex", "Install with: npm install -g @openai/codex or brew install --cask codex"),
 }
 var DroidConfig = &AppConfig{
 	Name:         "Droid",
 	DisplayName:  "Droid",
 	Command:      "droid",
 	EnvVars:      func(model string) []EnvVar { return nil },
 	Args:         func(model string) []string { return nil },
 	Setup:        setupDroidSettings,
 	CheckInstall: checkCommand("droid", "Install from: https://docs.factory.ai/cli/getting-started/quickstart"),
 }
 var AppRegistry = map[string]*AppConfig{
 	"claude":   ClaudeConfig,
 	"codex":    CodexConfig,
 	"droid":    DroidConfig,
 	"opencode": OpenCodeConfig,
 }
 func GetApp(name string) (*AppConfig, bool) {
 	app, ok := AppRegistry[strings.ToLower(name)]
 	return app, ok
 }
 func getModelInfo(model string) *api.ShowResponse {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return nil
 	}
 	resp, err := client.Show(context.Background(), &api.ShowRequest{Model: model})
 	if err != nil {
 		return nil
 	}
 	return resp
 }
 func getModelContextLength(model string) int {
 	const defaultCtx = 64000 // default context is set to 64k to support coding agents
 	resp := getModelInfo(model)
 	if resp == nil || resp.ModelInfo == nil {
 		return defaultCtx
 	}
 	arch, ok := resp.ModelInfo["general.architecture"].(string)
 	if !ok {
 		return defaultCtx
 	}
 	// currently being capped at 128k
 	if v, ok := resp.ModelInfo[fmt.Sprintf("%s.context_length", arch)].(float64); ok {
 		return min(int(v), 128000)
 	}
 	return defaultCtx
 }
 func modelSupportsImages(model string) bool {
 	resp := getModelInfo(model)
 	if resp == nil {
 		return false
 	}
 	return slices.Contains(resp.Capabilities, "vision")
 }
 func copyFile(src, dst string) error {
 	info, err := os.Stat(src)
 	if err != nil {
 		return err
 	}
 	data, err := os.ReadFile(src)
 	if err != nil {
 		return err
 	}
 	// Preserve source file permissions (important for files containing API keys)
 	return os.WriteFile(dst, data, info.Mode().Perm())
 }
 func atomicWriteJSON(path string, data any) error {
 	content, err := json.MarshalIndent(data, "", "  ")
 	if err != nil {
 		return fmt.Errorf("marshal failed: %w", err)
 	}
 	var check any
 	if err := json.Unmarshal(content, &check); err != nil {
 		return fmt.Errorf("validation failed: %w", err)
 	}
 	backupPath := path + ".bak"
 	if existingContent, err := os.ReadFile(path); err == nil {
 		if !bytes.Equal(existingContent, content) {
 			if err := copyFile(path, backupPath); err != nil {
 				return fmt.Errorf("backup failed: %w", err)
 			}
 		}
 	}
 	dir := filepath.Dir(path)
 	tmp, err := os.CreateTemp(dir, ".tmp-*")
 	if err != nil {
 		return fmt.Errorf("create temp failed: %w", err)
 	}
 	tmpPath := tmp.Name()
 	if _, err := tmp.Write(content); err != nil {
 		tmp.Close()
 		os.Remove(tmpPath)
 		return fmt.Errorf("write failed: %w", err)
 	}
 	if err := tmp.Close(); err != nil {
 		os.Remove(tmpPath)
 		return fmt.Errorf("close failed: %w", err)
 	}
 	if err := os.Rename(tmpPath, path); err != nil {
 		os.Remove(tmpPath)
 		if _, statErr := os.Stat(backupPath); statErr == nil {
 			os.Rename(backupPath, path)
 		}
 		return fmt.Errorf("rename failed: %w", err)
 	}
 	return nil
 }
 func isValidReasoningEffort(effort string) bool {
 	switch effort {
 	case "high", "medium", "low", "none":
 		return true
 	}
 	return false
 }
 func setupDroidSettings(models []string) error {
 	if len(models) == 0 {
 		return nil
 	}
 	home, err := os.UserHomeDir()
 	if err != nil {
 		return err
 	}
 	settingsPath := filepath.Join(home, ".factory", "settings.json")
 	if err := os.MkdirAll(filepath.Dir(settingsPath), 0755); err != nil {
 		return err
 	}
 	settings := make(map[string]any)
 	if data, err := os.ReadFile(settingsPath); err == nil {
 		json.Unmarshal(data, &settings)
 	}
 	var customModels []any
 	if existing, ok := settings["customModels"].([]any); ok {
 		customModels = existing
 	}
 	// Keep only non-Ollama models (we'll rebuild Ollama models fresh)
 	var nonOllamaModels []any
 	for _, m := range customModels {
 		entry, ok := m.(map[string]any)
 		if !ok {
 			nonOllamaModels = append(nonOllamaModels, m)
 			continue
 		}
 		displayName, _ := entry["displayName"].(string)
 		if !strings.HasSuffix(displayName, "[Ollama]") {
 			nonOllamaModels = append(nonOllamaModels, m)
 		}
 	}
 	// Build new Ollama model entries with sequential indices (0, 1, 2, ...)
 	var ollamaModels []any
 	var defaultModelID string
 	for i, model := range models {
 		modelID := fmt.Sprintf("custom:%s-[Ollama]-%d", model, i)
 		newEntry := map[string]any{
 			"model":           model,
 			"displayName":     fmt.Sprintf("%s [Ollama]", model),
 			"baseUrl":         "http://localhost:11434/v1",
 			"apiKey":          "ollama",
 			"provider":        "generic-chat-completion-api",
 			"maxOutputTokens": getModelContextLength(model),
 			"supportsImages":  modelSupportsImages(model),
 			"id":              modelID,
 			"index":           i,
 		}
 		ollamaModels = append(ollamaModels, newEntry)
 		if i == 0 {
 			defaultModelID = modelID
 		}
 	}
 	settings["customModels"] = append(ollamaModels, nonOllamaModels...)
 	sessionSettings, ok := settings["sessionDefaultSettings"].(map[string]any)
 	if !ok {
 		sessionSettings = make(map[string]any)
 	}
 	sessionSettings["model"] = defaultModelID
 	if effort, ok := sessionSettings["reasoningEffort"].(string); !ok || !isValidReasoningEffort(effort) {
 		sessionSettings["reasoningEffort"] = "none"
 	}
 	settings["sessionDefaultSettings"] = sessionSettings
 	return atomicWriteJSON(settingsPath, settings)
 }
 func setupOpenCodeSettings(modelList []string) error {
 	if len(modelList) == 0 {
 		return nil
 	}
 	home, err := os.UserHomeDir()
 	if err != nil {
 		return err
 	}
 	configPath := filepath.Join(home, ".config", "opencode", "opencode.json")
 	if err := os.MkdirAll(filepath.Dir(configPath), 0755); err != nil {
 		return err
 	}
 	config := make(map[string]any)
 	if data, err := os.ReadFile(configPath); err == nil {
 		json.Unmarshal(data, &config)
 	}
 	config["$schema"] = "https://opencode.ai/config.json"
 	provider, ok := config["provider"].(map[string]any)
 	if !ok {
 		provider = make(map[string]any)
 	}
 	ollama, ok := provider["ollama"].(map[string]any)
 	if !ok {
 		ollama = map[string]any{
 			"npm":  "@ai-sdk/openai-compatible",
 			"name": "Ollama (local)",
 			"options": map[string]any{
 				"baseURL": "http://localhost:11434/v1",
 			},
 		}
 	}
 	models, ok := ollama["models"].(map[string]any)
 	if !ok {
 		models = make(map[string]any)
 	}
 	selectedSet := make(map[string]bool)
 	for _, m := range modelList {
 		selectedSet[m] = true
 	}
 	for name, cfg := range models {
 		if cfgMap, ok := cfg.(map[string]any); ok {
 			if displayName, ok := cfgMap["name"].(string); ok {
 				if strings.HasSuffix(displayName, "[Ollama]") && !selectedSet[name] {
 					delete(models, name)
 				}
 			}
 		}
 	}
 	for _, model := range modelList {
 		models[model] = map[string]any{
 			"name": fmt.Sprintf("%s [Ollama]", model),
 		}
 	}
 	ollama["models"] = models
 	provider["ollama"] = ollama
 	config["provider"] = provider
 	if err := atomicWriteJSON(configPath, config); err != nil {
 		return err
 	}
 	statePath := filepath.Join(home, ".local", "state", "opencode", "model.json")
 	if err := os.MkdirAll(filepath.Dir(statePath), 0755); err != nil {
 		return err
 	}
 	state := map[string]any{
 		"recent":   []any{},
 		"favorite": []any{},
 		"variant":  map[string]any{},
 	}
 	if data, err := os.ReadFile(statePath); err == nil {
 		json.Unmarshal(data, &state)
 	}
 	recent, _ := state["recent"].([]any)
 	modelSet := make(map[string]bool)
 	for _, m := range modelList {
 		modelSet[m] = true
 	}
 	newRecent := []any{}
 	for _, entry := range recent {
 		if e, ok := entry.(map[string]any); ok {
 			if e["providerID"] == "ollama" {
 				if modelID, ok := e["modelID"].(string); ok && modelSet[modelID] {
 					continue
 				}
 			}
 		}
 		newRecent = append(newRecent, entry)
 	}
 	for i := len(modelList) - 1; i >= 0; i-- {
 		newRecent = append([]any{map[string]any{
 			"providerID": "ollama",
 			"modelID":    modelList[i],
 		}}, newRecent...)
 	}
 	if len(newRecent) > 10 {
 		newRecent = newRecent[:10]
 	}
 	state["recent"] = newRecent
 	return atomicWriteJSON(statePath, state)
 }
 var OpenCodeConfig = &AppConfig{
 	Name:         "OpenCode",
 	DisplayName:  "OpenCode",
 	Command:      "opencode",
 	EnvVars:      func(model string) []EnvVar { return nil },
 	Args:         func(model string) []string { return nil },
 	Setup:        setupOpenCodeSettings,
 	CheckInstall: checkCommand("opencode", "Install from: https://opencode.ai"),
 }
 func readJSONFile(path string) (map[string]any, error) {
 	data, err := os.ReadFile(path)
 	if err != nil {
 		return nil, err
 	}
 	var result map[string]any
 	if err := json.Unmarshal(data, &result); err != nil {
 		return nil, err
 	}
 	return result, nil
 }
 func getOpenCodeOllamaModels() (map[string]any, error) {
 	home, err := os.UserHomeDir()
 	if err != nil {
 		return nil, err
 	}
 	config, err := readJSONFile(filepath.Join(home, ".config", "opencode", "opencode.json"))
 	if err != nil {
 		return nil, err
 	}
 	provider, _ := config["provider"].(map[string]any)
 	ollama, _ := provider["ollama"].(map[string]any)
 	models, _ := ollama["models"].(map[string]any)
 	return models, nil
 }
 func getOpenCodeConfiguredModels() []string {
 	models, err := getOpenCodeOllamaModels()
 	if err != nil || models == nil {
 		return nil
 	}
 	var result []string
 	for name := range models {
 		result = append(result, name)
 	}
 	return result
 }
 func getOpenCodeConfiguredModel() string {
 	home, err := os.UserHomeDir()
 	if err != nil {
 		return ""
 	}
 	state, err := readJSONFile(filepath.Join(home, ".local", "state", "opencode", "model.json"))
 	if err != nil {
 		return ""
 	}
 	recent, _ := state["recent"].([]any)
 	if len(recent) == 0 {
 		return ""
 	}
 	first, _ := recent[0].(map[string]any)
 	if first["providerID"] == "ollama" {
 		modelID, _ := first["modelID"].(string)
 		return modelID
 	}
 	return ""
 }
 func readDroidSettings() (map[string]any, error) {
 	home, err := os.UserHomeDir()
 	if err != nil {
 		return nil, err
 	}
 	return readJSONFile(filepath.Join(home, ".factory", "settings.json"))
 }
 func getDroidConfiguredModels() []string {
 	settings, err := readDroidSettings()
 	if err != nil {
 		return nil
 	}
 	customModels, _ := settings["customModels"].([]any)
 	var result []string
 	for _, m := range customModels {
 		entry, _ := m.(map[string]any)
 		displayName, _ := entry["displayName"].(string)
 		// Only include Ollama models (those with our displayName pattern)
 		if strings.HasSuffix(displayName, "[Ollama]") {
 			if model, _ := entry["model"].(string); model != "" {
 				result = append(result, model)
 			}
 		}
 	}
 	return result
 }
 func getDroidConfiguredModel() string {
 	settings, err := readDroidSettings()
 	if err != nil {
 		return ""
 	}
 	sessionSettings, _ := settings["sessionDefaultSettings"].(map[string]any)
 	modelID, _ := sessionSettings["model"].(string)
 	if modelID == "" {
 		return ""
 	}
 	customModels, _ := settings["customModels"].([]any)
 	for _, m := range customModels {
 		entry, _ := m.(map[string]any)
 		if entry["id"] == modelID {
 			model, _ := entry["model"].(string)
 			return model
 		}
 	}
 	return ""
 }
 func getAppConfiguredModels(appName string) []string {
 	// Get models that exist in the app's config
 	var appModels []string
 	switch strings.ToLower(appName) {
 	case "opencode":
 		appModels = getOpenCodeConfiguredModels()
 	case "droid":
 		appModels = getDroidConfiguredModels()
 	}
 	// Get our saved connection config for the correct order (default first)
 	savedConfig, err := LoadConnection(appName)
 	if err != nil || len(savedConfig.Models) == 0 {
 		return appModels
 	}
 	if len(appModels) == 0 {
 		return savedConfig.Models
 	}
 	// Merge: saved order first (filtered to still-existing models), then any new models
 	appModelSet := make(map[string]bool, len(appModels))
 	for _, m := range appModels {
 		appModelSet[m] = true
 	}
 	seen := make(map[string]bool, len(savedConfig.Models))
 	var result []string
 	for _, m := range savedConfig.Models {
 		if appModelSet[m] {
 			result = append(result, m)
 			seen[m] = true
 		}
 	}
 	for _, m := range appModels {
 		if !seen[m] {
 			result = append(result, m)
 		}
 	}
 	return result
 }
 func printModelsAdded(appName string, models []string) {
 	if len(models) == 1 {
 		fmt.Fprintf(os.Stderr, "Added %s to %s\n", models[0], appName)
 	} else {
 		fmt.Fprintf(os.Stderr, "Added %d models to %s (default: %s)\n", len(models), appName, models[0])
 	}
 }
 func runInApp(appName, modelName string) error {
 	app, ok := GetApp(appName)
 	if !ok {
 		return fmt.Errorf("unknown app: %s", appName)
 	}
 	if err := app.CheckInstall(); err != nil {
 		return err
 	}
 	if app.Setup != nil {
 		models := []string{modelName}
 		if config, err := LoadConnection(appName); err == nil && len(config.Models) > 0 {
 			models = config.Models
 		}
 		if err := app.Setup(models); err != nil {
 			return fmt.Errorf("setup failed: %w", err)
 		}
 	}
 	proc := exec.Command(app.Command, app.Args(modelName)...)
 	proc.Stdin = os.Stdin
 	proc.Stdout = os.Stdout
 	proc.Stderr = os.Stderr
 	proc.Env = os.Environ()
 	for _, env := range app.EnvVars(modelName) {
 		proc.Env = append(proc.Env, fmt.Sprintf("%s=%s", env.Name, env.Value))
 	}
 	fmt.Fprintf(os.Stderr, "Launching %s with %s...\n", app.DisplayName, modelName)
 	return proc.Run()
 }
 // handleCancelled prints the cancellation message and returns true if err is ErrCancelled.
 // Returns false and the original error otherwise.
 func handleCancelled(err error) (cancelled bool, origErr error) {
 	if errors.Is(err, ErrCancelled) {
 		fmt.Fprintln(os.Stderr, err.Error())
 		return true, nil
 	}
 	return false, err
 }
 func ConnectCmd() *cobra.Command {
 	var modelFlag string
 	cmd := &cobra.Command{
 		Use:   "connect [APP]",
 		Short: "Configure an external app to use Ollama",
 		Long: `Configure an external application to use Ollama as its backend.
 Supported apps:
  claude    Claude Code
  codex     Codex
  droid     Droid
  opencode  OpenCode
 Examples:
  ollama connect
  ollama connect claude
  ollama connect claude --model llama3.2`,
 		Args:    cobra.MaximumNArgs(1),
 		PreRunE: checkServerHeartbeat,
 		RunE: func(cmd *cobra.Command, args []string) error {
 			var appName string
 			if len(args) > 0 {
 				appName = args[0]
 			} else {
 				var err error
 				appName, err = selectApp()
 				if cancelled, err := handleCancelled(err); cancelled {
 					return nil
 				} else if err != nil {
 					return err
 				}
 			}
 			if _, ok := GetApp(appName); !ok {
 				return fmt.Errorf("unknown app: %s", appName)
 			}
 			var models []string
 			if modelFlag != "" {
 				models = []string{modelFlag}
 			} else {
 				var err error
 				models, err = selectModelForConnect(cmd.Context(), appName)
 				if cancelled, err := handleCancelled(err); cancelled {
 					return nil
 				} else if err != nil {
 					return err
 				}
 			}
 			if err := SaveConnection(appName, models); err != nil {
 				return fmt.Errorf("failed to save: %w", err)
 			}
 			// Apply app-specific configuration (e.g., update OpenCode/Droid config files)
 			app, _ := GetApp(appName)
 			if app.Setup != nil {
 				if err := app.Setup(models); err != nil {
 					return fmt.Errorf("setup failed: %w", err)
 				}
 			}
 			printModelsAdded(appName, models)
 			if launch, _ := confirmLaunch(appName); launch {
 				return runInApp(appName, models[0])
 			}
 			fmt.Fprintf(os.Stderr, "Run 'ollama launch %s' to start later\n", strings.ToLower(appName))
 			return nil
 		},
 	}
 	cmd.Flags().StringVar(&modelFlag, "model", "", "Model to use")
 	return cmd
 }
 func getAppConfiguredModel(appName string) string {
 	switch strings.ToLower(appName) {
 	case "opencode":
 		return getOpenCodeConfiguredModel()
 	case "droid":
 		return getDroidConfiguredModel()
 	default:
 		return ""
 	}
 }
 func getOrConfigureModel(ctx context.Context, appName string) (string, error) {
 	if modelName := getAppConfiguredModel(appName); modelName != "" {
 		return modelName, nil
 	}
 	if config, err := LoadConnection(appName); err == nil {
 		return config.DefaultModel(), nil
 	} else if !os.IsNotExist(err) {
 		return "", err
 	}
 	models, err := selectModelForConnect(ctx, appName)
 	if err != nil {
 		return "", err
 	}
 	if err := SaveConnection(appName, models); err != nil {
 		return "", fmt.Errorf("failed to save: %w", err)
 	}
 	printModelsAdded(appName, models)
 	return models[0], nil
 }
 func LaunchCmd() *cobra.Command {
 	cmd := &cobra.Command{
 		Use:   "launch [APP]",
 		Short: "Launch a configured app",
 		Long: `Launch a configured application with Ollama as its backend.
 If no app is specified, shows a list of configured apps to choose from.
 If no apps have been configured, starts the connect flow.
 Examples:
  ollama launch
  ollama launch claude`,
 		Args:    cobra.MaximumNArgs(1),
 		PreRunE: checkServerHeartbeat,
 		RunE: func(cmd *cobra.Command, args []string) error {
 			var appName string
 			if len(args) > 0 {
 				appName = args[0]
 			} else {
 				selected, err := selectConnectedApp()
 				if errors.Is(err, ErrCancelled) {
 					return nil // Silent exit on cancel
 				} else if err != nil {
 					return err
 				}
 				if selected == "" {
 					fmt.Fprintf(os.Stderr, "No apps configured. Let's set one up.\n\n")
 					appName, err = selectApp()
 					if errors.Is(err, ErrCancelled) {
 						return nil
 					} else if err != nil {
 						return err
 					}
 				} else {
 					appName = selected
 				}
 			}
 			app, ok := GetApp(appName)
 			if !ok {
 				return fmt.Errorf("unknown app: %s", appName)
 			}
 			modelName, err := getOrConfigureModel(cmd.Context(), appName)
 			if errors.Is(err, ErrCancelled) {
 				return nil
 			} else if err != nil {
 				return err
 			}
 			return runInApp(app.Name, modelName)
 		},
 	}
 	return cmd
 }
--- a/cmd/apps_test.go
+++ b/cmd/apps_test.go
@@ -1,597 +0,0 @@
 package cmd
 import (
 	"encoding/json"
 	"os"
 	"path/filepath"
 	"testing"
 )
 func TestSetupOpenCodeSettings(t *testing.T) {
 	tmpDir := t.TempDir()
 	origHome := os.Getenv("HOME")
 	os.Setenv("HOME", tmpDir)
 	defer os.Setenv("HOME", origHome)
 	configDir := filepath.Join(tmpDir, ".config", "opencode")
 	configPath := filepath.Join(configDir, "opencode.json")
 	stateDir := filepath.Join(tmpDir, ".local", "state", "opencode")
 	statePath := filepath.Join(stateDir, "model.json")
 	cleanup := func() {
 		os.RemoveAll(configDir)
 		os.RemoveAll(stateDir)
 	}
 	t.Run("fresh install", func(t *testing.T) {
 		cleanup()
 		if err := setupOpenCodeSettings([]string{"llama3.2"}); err != nil {
 			t.Fatal(err)
 		}
 		assertOpenCodeModelExists(t, configPath, "llama3.2")
 		assertOpenCodeRecentModel(t, statePath, 0, "ollama", "llama3.2")
 	})
 	t.Run("preserve other providers", func(t *testing.T) {
 		cleanup()
 		os.MkdirAll(configDir, 0755)
 		os.WriteFile(configPath, []byte(`{"provider":{"anthropic":{"apiKey":"xxx"}}}`), 0644)
 		if err := setupOpenCodeSettings([]string{"llama3.2"}); err != nil {
 			t.Fatal(err)
 		}
 		data, _ := os.ReadFile(configPath)
 		var cfg map[string]any
 		json.Unmarshal(data, &cfg)
 		provider := cfg["provider"].(map[string]any)
 		if provider["anthropic"] == nil {
 			t.Error("anthropic provider was removed")
 		}
 		assertOpenCodeModelExists(t, configPath, "llama3.2")
 	})
 	t.Run("preserve other models", func(t *testing.T) {
 		cleanup()
 		os.MkdirAll(configDir, 0755)
 		os.WriteFile(configPath, []byte(`{"provider":{"ollama":{"models":{"mistral":{"name":"Mistral"}}}}}`), 0644)
 		if err := setupOpenCodeSettings([]string{"llama3.2"}); err != nil {
 			t.Fatal(err)
 		}
 		assertOpenCodeModelExists(t, configPath, "mistral")
 		assertOpenCodeModelExists(t, configPath, "llama3.2")
 	})
 	t.Run("update existing model", func(t *testing.T) {
 		cleanup()
 		setupOpenCodeSettings([]string{"llama3.2"})
 		setupOpenCodeSettings([]string{"llama3.2"})
 		assertOpenCodeModelExists(t, configPath, "llama3.2")
 	})
 	t.Run("preserve top-level keys", func(t *testing.T) {
 		cleanup()
 		os.MkdirAll(configDir, 0755)
 		os.WriteFile(configPath, []byte(`{"theme":"dark","keybindings":{}}`), 0644)
 		if err := setupOpenCodeSettings([]string{"llama3.2"}); err != nil {
 			t.Fatal(err)
 		}
 		data, _ := os.ReadFile(configPath)
 		var cfg map[string]any
 		json.Unmarshal(data, &cfg)
 		if cfg["theme"] != "dark" {
 			t.Error("theme was removed")
 		}
 		if cfg["keybindings"] == nil {
 			t.Error("keybindings was removed")
 		}
 	})
 	t.Run("model state - insert at index 0", func(t *testing.T) {
 		cleanup()
 		os.MkdirAll(stateDir, 0755)
 		os.WriteFile(statePath, []byte(`{"recent":[{"providerID":"anthropic","modelID":"claude"}],"favorite":[],"variant":{}}`), 0644)
 		if err := setupOpenCodeSettings([]string{"llama3.2"}); err != nil {
 			t.Fatal(err)
 		}
 		assertOpenCodeRecentModel(t, statePath, 0, "ollama", "llama3.2")
 		assertOpenCodeRecentModel(t, statePath, 1, "anthropic", "claude")
 	})
 	t.Run("model state - preserve favorites and variants", func(t *testing.T) {
 		cleanup()
 		os.MkdirAll(stateDir, 0755)
 		os.WriteFile(statePath, []byte(`{"recent":[],"favorite":[{"providerID":"x","modelID":"y"}],"variant":{"a":"b"}}`), 0644)
 		if err := setupOpenCodeSettings([]string{"llama3.2"}); err != nil {
 			t.Fatal(err)
 		}
 		data, _ := os.ReadFile(statePath)
 		var state map[string]any
 		json.Unmarshal(data, &state)
 		if len(state["favorite"].([]any)) != 1 {
 			t.Error("favorite was modified")
 		}
 		if state["variant"].(map[string]any)["a"] != "b" {
 			t.Error("variant was modified")
 		}
 	})
 	t.Run("model state - deduplicate on re-add", func(t *testing.T) {
 		cleanup()
 		os.MkdirAll(stateDir, 0755)
 		os.WriteFile(statePath, []byte(`{"recent":[{"providerID":"ollama","modelID":"llama3.2"},{"providerID":"anthropic","modelID":"claude"}],"favorite":[],"variant":{}}`), 0644)
 		if err := setupOpenCodeSettings([]string{"llama3.2"}); err != nil {
 			t.Fatal(err)
 		}
 		data, _ := os.ReadFile(statePath)
 		var state map[string]any
 		json.Unmarshal(data, &state)
 		recent := state["recent"].([]any)
 		if len(recent) != 2 {
 			t.Errorf("expected 2 recent entries, got %d", len(recent))
 		}
 		assertOpenCodeRecentModel(t, statePath, 0, "ollama", "llama3.2")
 	})
 	t.Run("remove model", func(t *testing.T) {
 		cleanup()
 		// First add two models
 		setupOpenCodeSettings([]string{"llama3.2", "mistral"})
 		assertOpenCodeModelExists(t, configPath, "llama3.2")
 		assertOpenCodeModelExists(t, configPath, "mistral")
 		// Then remove one by only selecting the other
 		setupOpenCodeSettings([]string{"llama3.2"})
 		assertOpenCodeModelExists(t, configPath, "llama3.2")
 		assertOpenCodeModelNotExists(t, configPath, "mistral")
 	})
 	t.Run("remove model preserves non-ollama models", func(t *testing.T) {
 		cleanup()
 		os.MkdirAll(configDir, 0755)
 		// Add a non-Ollama model manually
 		os.WriteFile(configPath, []byte(`{"provider":{"ollama":{"models":{"external":{"name":"External Model"}}}}}`), 0644)
 		setupOpenCodeSettings([]string{"llama3.2"})
 		assertOpenCodeModelExists(t, configPath, "llama3.2")
 		assertOpenCodeModelExists(t, configPath, "external") // Should be preserved
 	})
 }
 func assertOpenCodeModelExists(t *testing.T, path, model string) {
 	t.Helper()
 	data, err := os.ReadFile(path)
 	if err != nil {
 		t.Fatal(err)
 	}
 	var cfg map[string]any
 	if err := json.Unmarshal(data, &cfg); err != nil {
 		t.Fatal(err)
 	}
 	provider, ok := cfg["provider"].(map[string]any)
 	if !ok {
 		t.Fatal("provider not found")
 	}
 	ollama, ok := provider["ollama"].(map[string]any)
 	if !ok {
 		t.Fatal("ollama provider not found")
 	}
 	models, ok := ollama["models"].(map[string]any)
 	if !ok {
 		t.Fatal("models not found")
 	}
 	if models[model] == nil {
 		t.Errorf("model %s not found", model)
 	}
 }
 func assertOpenCodeModelNotExists(t *testing.T, path, model string) {
 	t.Helper()
 	data, err := os.ReadFile(path)
 	if err != nil {
 		t.Fatal(err)
 	}
 	var cfg map[string]any
 	if err := json.Unmarshal(data, &cfg); err != nil {
 		t.Fatal(err)
 	}
 	provider, ok := cfg["provider"].(map[string]any)
 	if !ok {
 		return // No provider means no model
 	}
 	ollama, ok := provider["ollama"].(map[string]any)
 	if !ok {
 		return // No ollama means no model
 	}
 	models, ok := ollama["models"].(map[string]any)
 	if !ok {
 		return // No models means no model
 	}
 	if models[model] != nil {
 		t.Errorf("model %s should not exist but was found", model)
 	}
 }
 func assertOpenCodeRecentModel(t *testing.T, path string, index int, providerID, modelID string) {
 	t.Helper()
 	data, err := os.ReadFile(path)
 	if err != nil {
 		t.Fatal(err)
 	}
 	var state map[string]any
 	if err := json.Unmarshal(data, &state); err != nil {
 		t.Fatal(err)
 	}
 	recent, ok := state["recent"].([]any)
 	if !ok {
 		t.Fatal("recent not found")
 	}
 	if index >= len(recent) {
 		t.Fatalf("index %d out of range (len=%d)", index, len(recent))
 	}
 	entry, ok := recent[index].(map[string]any)
 	if !ok {
 		t.Fatal("entry is not a map")
 	}
 	if entry["providerID"] != providerID {
 		t.Errorf("expected providerID %s, got %s", providerID, entry["providerID"])
 	}
 	if entry["modelID"] != modelID {
 		t.Errorf("expected modelID %s, got %s", modelID, entry["modelID"])
 	}
 }
 func TestSetupDroidSettings(t *testing.T) {
 	tmpDir := t.TempDir()
 	origHome := os.Getenv("HOME")
 	os.Setenv("HOME", tmpDir)
 	defer os.Setenv("HOME", origHome)
 	settingsDir := filepath.Join(tmpDir, ".factory")
 	settingsPath := filepath.Join(settingsDir, "settings.json")
 	cleanup := func() {
 		os.RemoveAll(settingsDir)
 	}
 	readSettings := func() map[string]any {
 		data, _ := os.ReadFile(settingsPath)
 		var settings map[string]any
 		json.Unmarshal(data, &settings)
 		return settings
 	}
 	getCustomModels := func(settings map[string]any) []map[string]any {
 		models, ok := settings["customModels"].([]any)
 		if !ok {
 			return nil
 		}
 		var result []map[string]any
 		for _, m := range models {
 			if entry, ok := m.(map[string]any); ok {
 				result = append(result, entry)
 			}
 		}
 		return result
 	}
 	t.Run("fresh install creates models with sequential indices", func(t *testing.T) {
 		cleanup()
 		if err := setupDroidSettings([]string{"model-a", "model-b"}); err != nil {
 			t.Fatal(err)
 		}
 		settings := readSettings()
 		models := getCustomModels(settings)
 		if len(models) != 2 {
 			t.Fatalf("expected 2 models, got %d", len(models))
 		}
 		// Check first model
 		if models[0]["model"] != "model-a" {
 			t.Errorf("expected model-a, got %s", models[0]["model"])
 		}
 		if models[0]["id"] != "custom:model-a-[Ollama]-0" {
 			t.Errorf("expected custom:model-a-[Ollama]-0, got %s", models[0]["id"])
 		}
 		if models[0]["index"] != float64(0) {
 			t.Errorf("expected index 0, got %v", models[0]["index"])
 		}
 		// Check second model
 		if models[1]["model"] != "model-b" {
 			t.Errorf("expected model-b, got %s", models[1]["model"])
 		}
 		if models[1]["id"] != "custom:model-b-[Ollama]-1" {
 			t.Errorf("expected custom:model-b-[Ollama]-1, got %s", models[1]["id"])
 		}
 		if models[1]["index"] != float64(1) {
 			t.Errorf("expected index 1, got %v", models[1]["index"])
 		}
 	})
 	t.Run("sets sessionDefaultSettings.model to first model ID", func(t *testing.T) {
 		cleanup()
 		if err := setupDroidSettings([]string{"model-a", "model-b"}); err != nil {
 			t.Fatal(err)
 		}
 		settings := readSettings()
 		session, ok := settings["sessionDefaultSettings"].(map[string]any)
 		if !ok {
 			t.Fatal("sessionDefaultSettings not found")
 		}
 		if session["model"] != "custom:model-a-[Ollama]-0" {
 			t.Errorf("expected custom:model-a-[Ollama]-0, got %s", session["model"])
 		}
 	})
 	t.Run("re-indexes when models removed", func(t *testing.T) {
 		cleanup()
 		// Add three models
 		setupDroidSettings([]string{"model-a", "model-b", "model-c"})
 		// Remove middle model
 		setupDroidSettings([]string{"model-a", "model-c"})
 		settings := readSettings()
 		models := getCustomModels(settings)
 		if len(models) != 2 {
 			t.Fatalf("expected 2 models, got %d", len(models))
 		}
 		// Check indices are sequential 0, 1
 		if models[0]["index"] != float64(0) {
 			t.Errorf("expected index 0, got %v", models[0]["index"])
 		}
 		if models[1]["index"] != float64(1) {
 			t.Errorf("expected index 1, got %v", models[1]["index"])
 		}
 		// Check IDs match new indices
 		if models[0]["id"] != "custom:model-a-[Ollama]-0" {
 			t.Errorf("expected custom:model-a-[Ollama]-0, got %s", models[0]["id"])
 		}
 		if models[1]["id"] != "custom:model-c-[Ollama]-1" {
 			t.Errorf("expected custom:model-c-[Ollama]-1, got %s", models[1]["id"])
 		}
 	})
 	t.Run("preserves non-Ollama custom models", func(t *testing.T) {
 		cleanup()
 		os.MkdirAll(settingsDir, 0755)
 		// Pre-existing non-Ollama model
 		os.WriteFile(settingsPath, []byte(`{
 			"customModels": [
 				{"model": "gpt-4", "displayName": "GPT-4", "provider": "openai"}
 			]
 		}`), 0644)
 		setupDroidSettings([]string{"model-a"})
 		settings := readSettings()
 		models := getCustomModels(settings)
 		if len(models) != 2 {
 			t.Fatalf("expected 2 models (1 Ollama + 1 non-Ollama), got %d", len(models))
 		}
 		// Ollama model should be first
 		if models[0]["model"] != "model-a" {
 			t.Errorf("expected Ollama model first, got %s", models[0]["model"])
 		}
 		// Non-Ollama model should be preserved at end
 		if models[1]["model"] != "gpt-4" {
 			t.Errorf("expected gpt-4 preserved, got %s", models[1]["model"])
 		}
 	})
 	t.Run("preserves other settings", func(t *testing.T) {
 		cleanup()
 		os.MkdirAll(settingsDir, 0755)
 		os.WriteFile(settingsPath, []byte(`{
 			"theme": "dark",
 			"enableHooks": true,
 			"sessionDefaultSettings": {"autonomyMode": "auto-high"}
 		}`), 0644)
 		setupDroidSettings([]string{"model-a"})
 		settings := readSettings()
 		if settings["theme"] != "dark" {
 			t.Error("theme was not preserved")
 		}
 		if settings["enableHooks"] != true {
 			t.Error("enableHooks was not preserved")
 		}
 		session := settings["sessionDefaultSettings"].(map[string]any)
 		if session["autonomyMode"] != "auto-high" {
 			t.Error("autonomyMode was not preserved")
 		}
 	})
 	t.Run("required fields present", func(t *testing.T) {
 		cleanup()
 		setupDroidSettings([]string{"test-model"})
 		settings := readSettings()
 		models := getCustomModels(settings)
 		if len(models) != 1 {
 			t.Fatal("expected 1 model")
 		}
 		model := models[0]
 		requiredFields := []string{"model", "displayName", "baseUrl", "apiKey", "provider", "maxOutputTokens", "id", "index"}
 		for _, field := range requiredFields {
 			if model[field] == nil {
 				t.Errorf("missing required field: %s", field)
 			}
 		}
 		if model["baseUrl"] != "http://localhost:11434/v1" {
 			t.Errorf("unexpected baseUrl: %s", model["baseUrl"])
 		}
 		if model["apiKey"] != "ollama" {
 			t.Errorf("unexpected apiKey: %s", model["apiKey"])
 		}
 		if model["provider"] != "generic-chat-completion-api" {
 			t.Errorf("unexpected provider: %s", model["provider"])
 		}
 	})
 	t.Run("fixes invalid reasoningEffort", func(t *testing.T) {
 		cleanup()
 		os.MkdirAll(settingsDir, 0755)
 		// Pre-existing settings with invalid reasoningEffort
 		os.WriteFile(settingsPath, []byte(`{
 			"sessionDefaultSettings": {"reasoningEffort": "off"}
 		}`), 0644)
 		setupDroidSettings([]string{"model-a"})
 		settings := readSettings()
 		session := settings["sessionDefaultSettings"].(map[string]any)
 		if session["reasoningEffort"] != "none" {
 			t.Errorf("expected reasoningEffort to be fixed to 'none', got %s", session["reasoningEffort"])
 		}
 	})
 	t.Run("preserves valid reasoningEffort", func(t *testing.T) {
 		cleanup()
 		os.MkdirAll(settingsDir, 0755)
 		os.WriteFile(settingsPath, []byte(`{
 			"sessionDefaultSettings": {"reasoningEffort": "high"}
 		}`), 0644)
 		setupDroidSettings([]string{"model-a"})
 		settings := readSettings()
 		session := settings["sessionDefaultSettings"].(map[string]any)
 		if session["reasoningEffort"] != "high" {
 			t.Errorf("expected reasoningEffort to remain 'high', got %s", session["reasoningEffort"])
 		}
 	})
 }
 func TestAtomicWriteJSON(t *testing.T) {
 	tmpDir := t.TempDir()
 	t.Run("creates file", func(t *testing.T) {
 		path := filepath.Join(tmpDir, "new.json")
 		data := map[string]string{"key": "value"}
 		if err := atomicWriteJSON(path, data); err != nil {
 			t.Fatal(err)
 		}
 		content, err := os.ReadFile(path)
 		if err != nil {
 			t.Fatal(err)
 		}
 		var result map[string]string
 		if err := json.Unmarshal(content, &result); err != nil {
 			t.Fatal(err)
 		}
 		if result["key"] != "value" {
 			t.Errorf("expected value, got %s", result["key"])
 		}
 	})
 	t.Run("creates backup", func(t *testing.T) {
 		path := filepath.Join(tmpDir, "backup.json")
 		backupPath := path + ".bak"
 		// Write initial file
 		os.WriteFile(path, []byte(`{"original": true}`), 0644)
 		// Update with atomicWriteJSON
 		if err := atomicWriteJSON(path, map[string]bool{"updated": true}); err != nil {
 			t.Fatal(err)
 		}
 		// Check backup exists with original content
 		backup, err := os.ReadFile(backupPath)
 		if err != nil {
 			t.Fatal("backup file not created")
 		}
 		var backupData map[string]bool
 		json.Unmarshal(backup, &backupData)
 		if !backupData["original"] {
 			t.Error("backup doesn't contain original data")
 		}
 		// Check new file has updated content
 		current, _ := os.ReadFile(path)
 		var currentData map[string]bool
 		json.Unmarshal(current, &currentData)
 		if !currentData["updated"] {
 			t.Error("file doesn't contain updated data")
 		}
 	})
 	t.Run("no backup for new file", func(t *testing.T) {
 		path := filepath.Join(tmpDir, "nobak.json")
 		backupPath := path + ".bak"
 		if err := atomicWriteJSON(path, map[string]string{"new": "file"}); err != nil {
 			t.Fatal(err)
 		}
 		if _, err := os.Stat(backupPath); !os.IsNotExist(err) {
 			t.Error("backup should not exist for new file")
 		}
 	})
 	t.Run("valid JSON output", func(t *testing.T) {
 		path := filepath.Join(tmpDir, "valid.json")
 		data := map[string]any{
 			"string": "hello",
 			"number": 42,
 			"nested": map[string]string{"a": "b"},
 		}
 		if err := atomicWriteJSON(path, data); err != nil {
 			t.Fatal(err)
 		}
 		content, _ := os.ReadFile(path)
 		var parsed map[string]any
 		if err := json.Unmarshal(content, &parsed); err != nil {
 			t.Errorf("output is not valid JSON: %v", err)
 		}
 	})
 	t.Run("no backup when content unchanged", func(t *testing.T) {
 		path := filepath.Join(tmpDir, "unchanged.json")
 		backupPath := path + ".bak"
 		data := map[string]string{"key": "value"}
 		// First write
 		if err := atomicWriteJSON(path, data); err != nil {
 			t.Fatal(err)
 		}
 		// Create a "stale" backup to verify it's not overwritten
 		os.WriteFile(backupPath, []byte(`{"stale": "backup"}`), 0644)
 		// Second write with same content
 		if err := atomicWriteJSON(path, data); err != nil {
 			t.Fatal(err)
 		}
 		// Backup should still contain stale content (not overwritten)
 		backup, _ := os.ReadFile(backupPath)
 		if string(backup) != `{"stale": "backup"}` {
 			t.Errorf("backup was overwritten when content unchanged")
 		}
 	})
 }
--- a/cmd/cmd.go
+++ b/cmd/cmd.go
@@ -1945,8 +1945,6 @@ func NewCLI() *cobra.Command {
 		copyCmd,
 		deleteCmd,
 		runnerCmd,
 		ConnectCmd(),
 		LaunchCmd(),
 	)
 	return rootCmd
--- a/cmd/config.go
+++ b/cmd/config.go
@@ -1,112 +0,0 @@
 package cmd
 import (
 	"encoding/json"
 	"os"
 	"path/filepath"
 	"strings"
 	"time"
 )
 type ConnectionConfig struct {
 	App          string    `json:"app"`
 	Models       []string  `json:"models"`
 	ConfiguredAt time.Time `json:"configured_at"`
 }
 // DefaultModel returns the first (default) model, or empty string if none
 func (c *ConnectionConfig) DefaultModel() string {
 	if len(c.Models) > 0 {
 		return c.Models[0]
 	}
 	return ""
 }
 func connectionsDir() (string, error) {
 	home, err := os.UserHomeDir()
 	if err != nil {
 		return "", err
 	}
 	return filepath.Join(home, ".ollama", "connections"), nil
 }
 func configPath(appName string) (string, error) {
 	dir, err := connectionsDir()
 	if err != nil {
 		return "", err
 	}
 	// Normalize to lowercase for consistent file naming across case-sensitive filesystems
 	return filepath.Join(dir, strings.ToLower(appName)+".json"), nil
 }
 func SaveConnection(appName string, models []string) error {
 	path, err := configPath(appName)
 	if err != nil {
 		return err
 	}
 	if err := os.MkdirAll(filepath.Dir(path), 0755); err != nil {
 		return err
 	}
 	return atomicWriteJSON(path, ConnectionConfig{
 		App:          appName,
 		Models:       models,
 		ConfiguredAt: time.Now(),
 	})
 }
 func LoadConnection(appName string) (*ConnectionConfig, error) {
 	path, err := configPath(appName)
 	if err != nil {
 		return nil, err
 	}
 	data, err := os.ReadFile(path)
 	if err != nil {
 		return nil, err
 	}
 	var config ConnectionConfig
 	if err := json.Unmarshal(data, &config); err != nil {
 		return nil, err
 	}
 	return &config, nil
 }
 func ListConnections() ([]ConnectionConfig, error) {
 	dir, err := connectionsDir()
 	if err != nil {
 		return nil, err
 	}
 	entries, err := os.ReadDir(dir)
 	if err != nil {
 		if os.IsNotExist(err) {
 			return nil, nil
 		}
 		return nil, err
 	}
 	var configs []ConnectionConfig
 	for _, entry := range entries {
 		if entry.IsDir() || filepath.Ext(entry.Name()) != ".json" {
 			continue
 		}
 		data, err := os.ReadFile(filepath.Join(dir, entry.Name()))
 		if err != nil {
 			continue
 		}
 		var config ConnectionConfig
 		if err := json.Unmarshal(data, &config); err != nil {
 			continue
 		}
 		configs = append(configs, config)
 	}
 	return configs, nil
 }
--- a/cmd/interactive.go
+++ b/cmd/interactive.go
@@ -37,8 +37,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 		fmt.Fprintln(os.Stderr, "  /load <model>   Load a session or model")
 		fmt.Fprintln(os.Stderr, "  /save <model>   Save your current session")
 		fmt.Fprintln(os.Stderr, "  /clear          Clear session context")
 		fmt.Fprintln(os.Stderr, "  /connect        Configure an external app to use Ollama")
 		fmt.Fprintln(os.Stderr, "  /launch [app]   Launch a configured app")
 		fmt.Fprintln(os.Stderr, "  /bye            Exit")
 		fmt.Fprintln(os.Stderr, "  /?, /help       Help for a command")
 		fmt.Fprintln(os.Stderr, "  /? shortcuts    Help for keyboard shortcuts")
@@ -462,104 +460,6 @@ func generateInteractive(cmd *cobra.Command, opts runOptions) error {
 			}
 		case strings.HasPrefix(line, "/exit"), strings.HasPrefix(line, "/bye"):
 			return nil
 		case strings.HasPrefix(line, "/connect"):
 			args := strings.Fields(line)
 			var appName string
 			if len(args) > 1 {
 				appName = args[1]
 			} else {
 				var err error
 				appName, err = selectApp()
 				if cancelled, _ := handleCancelled(err); cancelled {
 					continue
 				} else if err != nil {
 					fmt.Printf("error: %v\n", err)
 					continue
 				}
 			}
 			if _, ok := GetApp(appName); !ok {
 				fmt.Printf("Unknown app: %s\n", appName)
 				continue
 			}
 			models, err := selectModelForConnect(cmd.Context(), appName)
 			if cancelled, _ := handleCancelled(err); cancelled {
 				continue
 			} else if err != nil {
 				fmt.Printf("error: %v\n", err)
 				continue
 			}
 			if err := SaveConnection(appName, models); err != nil {
 				fmt.Printf("error: %v\n", err)
 				continue
 			}
 			printModelsAdded(appName, models)
 			if launch, _ := confirmLaunch(appName); launch {
 				if err := runInApp(appName, models[0]); err != nil {
 					fmt.Printf("error: %v\n", err)
 				}
 			}
 			continue
 		case strings.HasPrefix(line, "/launch"):
 			args := strings.Fields(line)
 			var appName string
 			if len(args) >= 2 {
 				appName = args[1]
 			} else {
 				selected, err := selectConnectedApp()
 				if errors.Is(err, ErrCancelled) {
 					continue // Silent exit on cancel
 				} else if err != nil {
 					fmt.Printf("error: %v\n", err)
 					continue
 				}
 				if selected == "" {
 					fmt.Fprintf(os.Stderr, "No apps configured. Let's set one up.\n\n")
 					appName, err = selectApp()
 					if errors.Is(err, ErrCancelled) {
 						continue
 					} else if err != nil {
 						fmt.Printf("error: %v\n", err)
 						continue
 					}
 				} else {
 					appName = selected
 				}
 			}
 			app, ok := GetApp(appName)
 			if !ok {
 				fmt.Printf("Unknown app: %s\n", appName)
 				continue
 			}
 			modelName, err := getOrConfigureModel(cmd.Context(), appName)
 			if errors.Is(err, ErrCancelled) {
 				continue
 			} else if err != nil {
 				fmt.Printf("error: %v\n", err)
 				continue
 			}
 			if opts.Model != "" && modelName != opts.Model {
 				if switchModel, _ := confirmPrompt(fmt.Sprintf("Switch %s to use %s?", app.DisplayName, opts.Model)); switchModel {
 					modelName = opts.Model
 					if err := SaveConnection(appName, []string{modelName}); err != nil {
 						fmt.Printf("error: %v\n", err)
 						continue
 					}
 					fmt.Fprintf(os.Stderr, "Updated %s to %s\n", appName, modelName)
 				}
 			}
 			if err := runInApp(app.Name, modelName); err != nil {
 				fmt.Printf("error: %v\n", err)
 			}
 			continue
 		case strings.HasPrefix(line, "/"):
 			args := strings.Fields(line)
 			isFile := false
--- a/cmd/selector.go
+++ b/cmd/selector.go
@@ -1,562 +0,0 @@
 package cmd
 import (
 	"context"
 	"errors"
 	"fmt"
 	"os"
 	"sort"
 	"strings"
 	"time"
 	"golang.org/x/term"
 	"github.com/ollama/ollama/api"
 )
 const maxDisplayedItems = 10
 var AppOrder = []string{"claude", "codex", "opencode", "droid"}
 var ErrCancelled = errors.New("no ollama connections created")
 type SelectItem struct {
 	Name        string
 	Description string
 }
 // terminalState manages raw terminal mode for interactive selection
 type terminalState struct {
 	fd       int
 	oldState *term.State
 }
 // enterRawMode puts terminal in raw mode with cursor hidden
 func enterRawMode() (*terminalState, error) {
 	fd := int(os.Stdin.Fd())
 	oldState, err := term.MakeRaw(fd)
 	if err != nil {
 		return nil, err
 	}
 	fmt.Fprint(os.Stderr, "\033[?25l") // hide cursor
 	return &terminalState{fd: fd, oldState: oldState}, nil
 }
 // restore restores terminal state and shows cursor
 func (t *terminalState) restore() {
 	fmt.Fprint(os.Stderr, "\033[?25h") // show cursor
 	term.Restore(t.fd, t.oldState)
 }
 // clearLines moves cursor up n lines and clears from there
 func clearLines(n int) {
 	if n > 0 {
 		fmt.Fprintf(os.Stderr, "\033[%dA", n)
 		fmt.Fprint(os.Stderr, "\033[J")
 	}
 }
 func Select(prompt string, items []SelectItem) (string, error) {
 	if len(items) == 0 {
 		return "", fmt.Errorf("no items to select from")
 	}
 	ts, err := enterRawMode()
 	if err != nil {
 		return "", err
 	}
 	defer ts.restore()
 	var filter string
 	selected := 0
 	scrollOffset := 0
 	var lastLineCount int
 	render := func() {
 		filtered := filterItems(items, filter)
 		clearLines(lastLineCount)
 		fmt.Fprintf(os.Stderr, "%s %s\r\n", prompt, filter)
 		lineCount := 1
 		if len(filtered) == 0 {
 			fmt.Fprintf(os.Stderr, "  \033[37m(no matches)\033[0m\r\n")
 			lineCount++
 		} else {
 			displayCount := min(len(filtered), maxDisplayedItems)
 			for i := range displayCount {
 				idx := scrollOffset + i
 				if idx >= len(filtered) {
 					break
 				}
 				item := filtered[idx]
 				prefix := "    "
 				if idx == selected {
 					prefix = "  \033[1m> "
 				}
 				if item.Description != "" {
 					fmt.Fprintf(os.Stderr, "%s%s\033[0m \033[37m- %s\033[0m\r\n", prefix, item.Name, item.Description)
 				} else {
 					fmt.Fprintf(os.Stderr, "%s%s\033[0m\r\n", prefix, item.Name)
 				}
 				lineCount++
 			}
 			if remaining := len(filtered) - scrollOffset - displayCount; remaining > 0 {
 				fmt.Fprintf(os.Stderr, "  \033[37m... and %d more\033[0m\r\n", remaining)
 				lineCount++
 			}
 		}
 		lastLineCount = lineCount
 	}
 	render()
 	buf := make([]byte, 3)
 	for {
 		n, err := os.Stdin.Read(buf)
 		if err != nil {
 			return "", err
 		}
 		filtered := filterItems(items, filter)
 		switch {
 		case n == 1 && buf[0] == 13: // Enter
 			if len(filtered) > 0 && selected < len(filtered) {
 				clearLines(lastLineCount)
 				return filtered[selected].Name, nil
 			}
 		case n == 1 && (buf[0] == 3 || buf[0] == 27): // Ctrl+C or Escape
 			clearLines(lastLineCount)
 			return "", ErrCancelled
 		case n == 1 && buf[0] == 127: // Backspace
 			if len(filter) > 0 {
 				filter = filter[:len(filter)-1]
 				selected = 0
 				scrollOffset = 0
 			}
 		case n == 3 && buf[0] == 27 && buf[1] == 91: // Arrow keys
 			if buf[2] == 65 && selected > 0 { // Up
 				selected--
 				if selected < scrollOffset {
 					scrollOffset = selected
 				}
 			} else if buf[2] == 66 && selected < len(filtered)-1 { // Down
 				selected++
 				if selected >= scrollOffset+maxDisplayedItems {
 					scrollOffset = selected - maxDisplayedItems + 1
 				}
 			}
 		case n == 1 && buf[0] >= 32 && buf[0] < 127: // Printable chars
 			filter += string(buf[0])
 			selected = 0
 			scrollOffset = 0
 		}
 		render()
 	}
 }
 func filterItems(items []SelectItem, filter string) []SelectItem {
 	if filter == "" {
 		return items
 	}
 	var result []SelectItem
 	filterLower := strings.ToLower(filter)
 	for _, item := range items {
 		if strings.Contains(strings.ToLower(item.Name), filterLower) {
 			result = append(result, item)
 		}
 	}
 	return result
 }
 func MultiSelect(prompt string, items []SelectItem, preChecked []string) ([]string, error) {
 	if len(items) == 0 {
 		return nil, fmt.Errorf("no items to select from")
 	}
 	ts, err := enterRawMode()
 	if err != nil {
 		return nil, err
 	}
 	defer ts.restore()
 	var filter string
 	highlighted := 0
 	scrollOffset := 0
 	checked := make(map[int]bool)
 	var checkOrder []int
 	var lastLineCount int
 	focusOnButton := false
 	// Build index lookup for O(1) access
 	itemIndex := make(map[string]int, len(items))
 	for i, item := range items {
 		itemIndex[item.Name] = i
 	}
 	// Pre-check items in their original order (preserves default as first)
 	for _, name := range preChecked {
 		if idx, ok := itemIndex[name]; ok {
 			checked[idx] = true
 			checkOrder = append(checkOrder, idx)
 		}
 	}
 	render := func() {
 		filtered := filterItems(items, filter)
 		clearLines(lastLineCount)
 		fmt.Fprintf(os.Stderr, "%s %s\r\n", prompt, filter)
 		lineCount := 1
 		if len(filtered) == 0 {
 			fmt.Fprintf(os.Stderr, "  \033[37m(no matches)\033[0m\r\n")
 			lineCount++
 		} else {
 			displayCount := min(len(filtered), maxDisplayedItems)
 			for i := range displayCount {
 				idx := scrollOffset + i
 				if idx >= len(filtered) {
 					break
 				}
 				item := filtered[idx]
 				origIdx := itemIndex[item.Name]
 				checkbox := "[ ]"
 				if checked[origIdx] {
 					checkbox = "[x]"
 				}
 				prefix := "  "
 				suffix := ""
 				if idx == highlighted && !focusOnButton {
 					prefix = "> "
 				}
 				if len(checkOrder) > 0 && checkOrder[0] == origIdx {
 					suffix = " \033[37m(default)\033[0m"
 				}
 				if idx == highlighted && !focusOnButton {
 					fmt.Fprintf(os.Stderr, "  \033[1m%s %s %s\033[0m%s\r\n", prefix, checkbox, item.Name, suffix)
 				} else {
 					fmt.Fprintf(os.Stderr, "  %s %s %s%s\r\n", prefix, checkbox, item.Name, suffix)
 				}
 				lineCount++
 			}
 			if remaining := len(filtered) - scrollOffset - displayCount; remaining > 0 {
 				fmt.Fprintf(os.Stderr, "  \033[37m... and %d more\033[0m\r\n", remaining)
 				lineCount++
 			}
 		}
 		// Continue button
 		fmt.Fprintf(os.Stderr, "\r\n")
 		lineCount++
 		count := len(checkOrder)
 		switch {
 		case count == 0:
 			fmt.Fprintf(os.Stderr, "  \033[37mSelect at least one model.\033[0m\r\n")
 		case focusOnButton:
 			fmt.Fprintf(os.Stderr, "  \033[1m> [ Continue ]\033[0m \033[37m(%d selected)\033[0m\r\n", count)
 		default:
 			fmt.Fprintf(os.Stderr, "    \033[37m[ Continue ] (%d selected) - press Tab\033[0m\r\n", count)
 		}
 		lineCount++
 		lastLineCount = lineCount
 	}
 	toggleItem := func() {
 		filtered := filterItems(items, filter)
 		if len(filtered) == 0 || highlighted >= len(filtered) {
 			return
 		}
 		item := filtered[highlighted]
 		origIdx := itemIndex[item.Name]
 		if checked[origIdx] {
 			delete(checked, origIdx)
 			for i, idx := range checkOrder {
 				if idx == origIdx {
 					checkOrder = append(checkOrder[:i], checkOrder[i+1:]...)
 					break
 				}
 			}
 		} else {
 			checked[origIdx] = true
 			checkOrder = append(checkOrder, origIdx)
 		}
 	}
 	render()
 	buf := make([]byte, 3)
 	for {
 		n, err := os.Stdin.Read(buf)
 		if err != nil {
 			return nil, err
 		}
 		filtered := filterItems(items, filter)
 		switch {
 		case n == 1 && buf[0] == 13: // Enter
 			if focusOnButton && len(checkOrder) > 0 {
 				clearLines(lastLineCount)
 				var result []string
 				for _, idx := range checkOrder {
 					result = append(result, items[idx].Name)
 				}
 				return result, nil
 			} else if !focusOnButton {
 				toggleItem()
 			}
 		case n == 1 && buf[0] == 9: // Tab
 			if len(checkOrder) > 0 {
 				focusOnButton = !focusOnButton
 			}
 		case n == 1 && (buf[0] == 3 || buf[0] == 27): // Ctrl+C or Escape
 			clearLines(lastLineCount)
 			return nil, ErrCancelled
 		case n == 1 && buf[0] == 127: // Backspace
 			if len(filter) > 0 {
 				filter = filter[:len(filter)-1]
 				highlighted = 0
 				scrollOffset = 0
 				focusOnButton = false
 			}
 		case n == 3 && buf[0] == 27 && buf[1] == 91: // Arrow keys
 			if focusOnButton {
 				// Any arrow key returns focus to list
 				focusOnButton = false
 			} else {
 				if buf[2] == 65 && highlighted > 0 { // Up
 					highlighted--
 					if highlighted < scrollOffset {
 						scrollOffset = highlighted
 					}
 				} else if buf[2] == 66 && highlighted < len(filtered)-1 { // Down
 					highlighted++
 					if highlighted >= scrollOffset+maxDisplayedItems {
 						scrollOffset = highlighted - maxDisplayedItems + 1
 					}
 				}
 			}
 		case n == 1 && buf[0] >= 32 && buf[0] < 127: // Printable chars
 			filter += string(buf[0])
 			highlighted = 0
 			scrollOffset = 0
 			focusOnButton = false
 		}
 		render()
 	}
 }
 func selectApp() (string, error) {
 	var items []SelectItem
 	for _, name := range AppOrder {
 		app, ok := AppRegistry[name]
 		if !ok {
 			continue
 		}
 		description := app.DisplayName
 		// Show configured model if one exists
 		if conn, err := LoadConnection(name); err == nil && conn.DefaultModel() != "" {
 			description = fmt.Sprintf("%s (%s)", app.DisplayName, conn.DefaultModel())
 		}
 		items = append(items, SelectItem{Name: app.Name, Description: description})
 	}
 	if len(items) == 0 {
 		return "", fmt.Errorf("no apps available")
 	}
 	return Select("Select app:", items)
 }
 func selectConnectedApp() (string, error) {
 	connections, err := ListConnections()
 	if err != nil {
 		return "", err
 	}
 	if len(connections) == 0 {
 		return "", nil
 	}
 	var items []SelectItem
 	for _, conn := range connections {
 		app, ok := GetApp(conn.App)
 		if !ok {
 			continue
 		}
 		items = append(items, SelectItem{
 			Name:        app.Name,
 			Description: fmt.Sprintf("%s (%s)", app.DisplayName, conn.DefaultModel()),
 		})
 	}
 	if len(items) == 0 {
 		return "", nil
 	}
 	return Select("Select app to launch (use ollama connect to configure other apps):", items)
 }
 func confirmLaunch(appName string) (bool, error) {
 	return confirmPrompt(fmt.Sprintf("Launch %s now?", appName))
 }
 func confirmPrompt(prompt string) (bool, error) {
 	fd := int(os.Stdin.Fd())
 	oldState, err := term.MakeRaw(fd)
 	if err != nil {
 		return false, err
 	}
 	defer term.Restore(fd, oldState)
 	fmt.Fprintf(os.Stderr, "%s [y/n] ", prompt)
 	buf := make([]byte, 1)
 	for {
 		if _, err := os.Stdin.Read(buf); err != nil {
 			return false, err
 		}
 		switch buf[0] {
 		case 'Y', 'y', 13: // Yes, Enter
 			fmt.Fprintf(os.Stderr, "yes\r\n")
 			return true, nil
 		case 'N', 'n', 27, 3: // No, Escape, Ctrl+C
 			fmt.Fprintf(os.Stderr, "no\r\n")
 			return false, nil
 		}
 	}
 }
 func selectModelForConnect(ctx context.Context, appName string) ([]string, error) {
 	client, err := api.ClientFromEnvironment()
 	if err != nil {
 		return nil, err
 	}
 	models, err := client.List(ctx)
 	if err != nil {
 		return nil, err
 	}
 	if len(models.Models) == 0 {
 		return nil, fmt.Errorf("no models available. Run 'ollama pull <model>' first")
 	}
 	var items []SelectItem
 	cloudModels := make(map[string]bool)
 	for _, m := range models.Models {
 		if m.RemoteModel != "" {
 			cloudModels[m.Name] = true
 		}
 		items = append(items, SelectItem{Name: m.Name})
 	}
 	if len(items) == 0 {
 		return nil, fmt.Errorf("no local models available. Run 'ollama pull <model>' first")
 	}
 	// Get already configured models for this app to pre-check
 	preChecked := getAppConfiguredModels(appName)
 	preCheckedSet := make(map[string]bool)
 	for _, name := range preChecked {
 		preCheckedSet[name] = true
 	}
 	sort.Slice(items, func(i, j int) bool {
 		iName, jName := strings.ToLower(items[i].Name), strings.ToLower(items[j].Name)
 		iChecked, jChecked := preCheckedSet[items[i].Name], preCheckedSet[items[j].Name]
 		// Pre-checked models come first
 		if iChecked != jChecked {
 			return iChecked
 		}
 		// Within each group, sort alphabetically
 		return iName < jName
 	})
 	// Apps with config files support multi-model, others use single-select
 	app, _ := GetApp(appName)
 	supportsMultiModel := app != nil && app.Setup != nil
 	var selected []string
 	if supportsMultiModel {
 		selected, err = MultiSelect(fmt.Sprintf("Select models for %s:", app.DisplayName), items, preChecked)
 		if err != nil {
 			return nil, err
 		}
 	} else {
 		model, err := Select(fmt.Sprintf("Select model for %s:", app.DisplayName), items)
 		if err != nil {
 			return nil, err
 		}
 		selected = []string{model}
 	}
 	// Check if any selected model is cloud, ensure signed in once
 	for _, model := range selected {
 		if cloudModels[model] {
 			if err := ensureSignedIn(ctx, client); err != nil {
 				return nil, err
 			}
 			break
 		}
 	}
 	return selected, nil
 }
 func ensureSignedIn(ctx context.Context, client *api.Client) error {
 	user, err := client.Whoami(ctx)
 	if err == nil && user != nil && user.Name != "" {
 		return nil
 	}
 	var aErr api.AuthorizationError
 	if !errors.As(err, &aErr) || aErr.SigninURL == "" {
 		return err
 	}
 	yes, err := confirmPrompt("Sign in to ollama.com?")
 	if err != nil || !yes {
 		return fmt.Errorf("sign in required for cloud models")
 	}
 	fmt.Fprintf(os.Stderr, "\nTo sign in, navigate to:\n    %s\n\n", aErr.SigninURL)
 	fmt.Fprintf(os.Stderr, "\033[90mwaiting for sign in to complete...\033[0m")
 	ticker := time.NewTicker(2 * time.Second)
 	defer ticker.Stop()
 	for {
 		select {
 		case <-ctx.Done():
 			fmt.Fprintf(os.Stderr, "\n")
 			return ctx.Err()
 		case <-ticker.C:
 			user, err := client.Whoami(ctx)
 			if err == nil && user != nil && user.Name != "" {
 				fmt.Fprintf(os.Stderr, "\r\033[K\033[A\r\033[K\033[1msigned in:\033[0m %s\n", user.Name)
 				return nil
 			}
 			fmt.Fprintf(os.Stderr, ".")
 		}
 	}
 }
--- a/scripts/build_darwin.sh
+++ b/scripts/build_darwin.sh
@@ -73,7 +73,7 @@ _build_darwin() {
            MLX_CGO_CFLAGS="-O3 -I$(pwd)/$BUILD_DIR/_deps/mlx-c-src -mmacosx-version-min=14.0"
            MLX_CGO_LDFLAGS="-L$(pwd)/$BUILD_DIR/lib/ollama -lmlxc -lmlx -Wl,-rpath,@executable_path -lc++ -framework Metal -framework Foundation -framework Accelerate -mmacosx-version-min=14.0"
        fi
-        GOOS=darwin GOARCH=$ARCH CGO_ENABLED=1 CGO_CFLAGS="$MLX_CGO_CFLAGS" CGO_LDFLAGS="$MLX_CGO_LDFLAGS" go build -tags mlx -o $INSTALL_PREFIX/imagegen ./x/imagegen/cmd/engine
+        GOOS=darwin GOARCH=$ARCH CGO_ENABLED=1 CGO_CFLAGS="$MLX_CGO_CFLAGS" CGO_LDFLAGS="$MLX_CGO_LDFLAGS" go build -tags mlx -o $INSTALL_PREFIX/ollama-mlx .
        GOOS=darwin GOARCH=$ARCH CGO_ENABLED=1 go build -o $INSTALL_PREFIX .
    done
 }
@@ -82,19 +82,19 @@ _sign_darwin() {
    status "Creating universal binary..."
    mkdir -p dist/darwin
    lipo -create -output dist/darwin/ollama dist/darwin-*/ollama
-    lipo -create -output dist/darwin/imagegen dist/darwin-*/imagegen
+    lipo -create -output dist/darwin/ollama-mlx dist/darwin-*/ollama-mlx
    chmod +x dist/darwin/ollama
-    chmod +x dist/darwin/imagegen
+    chmod +x dist/darwin/ollama-mlx
    if [ -n "$APPLE_IDENTITY" ]; then
-        for F in dist/darwin/ollama dist/darwin-*/lib/ollama/* dist/darwin/imagegen; do
+        for F in dist/darwin/ollama dist/darwin-*/lib/ollama/* dist/darwin/ollama-mlx; do
            codesign -f --timestamp -s "$APPLE_IDENTITY" --identifier ai.ollama.ollama --options=runtime $F
        done
        # create a temporary zip for notarization
        TEMP=$(mktemp -u).zip
        ditto -c -k --keepParent dist/darwin/ollama "$TEMP"
-        xcrun notarytool submit "$TEMP" --wait --timeout 10m --apple-id $APPLE_ID --password $APPLE_PASSWORD --team-id $APPLE_TEAM_ID
+        xcrun notarytool submit "$TEMP" --wait --timeout 20m --apple-id $APPLE_ID --password $APPLE_PASSWORD --team-id $APPLE_TEAM_ID
        rm -f "$TEMP"
    fi
@@ -154,23 +154,25 @@ _build_macapp() {
    mkdir -p dist/Ollama.app/Contents/Resources
    if [ -d dist/darwin-amd64 ]; then
        lipo -create -output dist/Ollama.app/Contents/Resources/ollama dist/darwin-amd64/ollama dist/darwin-arm64/ollama
-        lipo -create -output dist/Ollama.app/Contents/Resources/imagegen dist/darwin-amd64/imagegen dist/darwin-arm64/imagegen
+        lipo -create -output dist/Ollama.app/Contents/Resources/ollama-mlx dist/darwin-amd64/ollama-mlx dist/darwin-arm64/ollama-mlx
        for F in dist/darwin-amd64/lib/ollama/*mlx*.dylib ; do
            lipo -create -output dist/darwin/$(basename $F) $F dist/darwin-arm64/lib/ollama/$(basename $F)
        done
        cp dist/darwin-*/lib/ollama/*.so dist/darwin-*/lib/ollama/*.dylib dist/Ollama.app/Contents/Resources/
        cp dist/darwin/*.dylib dist/Ollama.app/Contents/Resources/
        # Copy MLX metallib (architecture-independent, just use arm64 version)
        cp dist/darwin-arm64/lib/ollama/*.metallib dist/Ollama.app/Contents/Resources/ 2>/dev/null || true
    else
        cp -a dist/darwin/ollama dist/Ollama.app/Contents/Resources/ollama
        cp dist/darwin/*.so dist/darwin/*.dylib dist/Ollama.app/Contents/Resources/
    fi
-    cp -a dist/darwin/imagegen dist/Ollama.app/Contents/Resources/imagegen
+    cp -a dist/darwin/ollama-mlx dist/Ollama.app/Contents/Resources/ollama-mlx
    chmod a+x dist/Ollama.app/Contents/Resources/ollama
    # Sign
    if [ -n "$APPLE_IDENTITY" ]; then
        codesign -f --timestamp -s "$APPLE_IDENTITY" --identifier ai.ollama.ollama --options=runtime dist/Ollama.app/Contents/Resources/ollama
-        for lib in dist/Ollama.app/Contents/Resources/*.so dist/Ollama.app/Contents/Resources/*.dylib dist/Ollama.app/Contents/Resources/imagegen ; do
+        for lib in dist/Ollama.app/Contents/Resources/*.so dist/Ollama.app/Contents/Resources/*.dylib dist/Ollama.app/Contents/Resources/*.metallib dist/Ollama.app/Contents/Resources/ollama-mlx ; do
            codesign -f --timestamp -s "$APPLE_IDENTITY" --identifier ai.ollama.ollama --options=runtime ${lib}
        done
        codesign -f --timestamp -s "$APPLE_IDENTITY" --identifier com.electron.ollama --deep --options=runtime dist/Ollama.app
@@ -178,11 +180,11 @@ _build_macapp() {
    rm -f dist/Ollama-darwin.zip
    ditto -c -k --keepParent dist/Ollama.app dist/Ollama-darwin.zip
-    (cd dist/Ollama.app/Contents/Resources/; tar -cf - ollama imagegen *.so *.dylib) | gzip -9vc > dist/ollama-darwin.tgz
+    (cd dist/Ollama.app/Contents/Resources/; tar -cf - ollama ollama-mlx *.so *.dylib *.metallib 2>/dev/null) | gzip -9vc > dist/ollama-darwin.tgz
    # Notarize and Staple
    if [ -n "$APPLE_IDENTITY" ]; then
-        $(xcrun -f notarytool) submit dist/Ollama-darwin.zip --wait --timeout 10m --apple-id "$APPLE_ID" --password "$APPLE_PASSWORD" --team-id "$APPLE_TEAM_ID"
+        $(xcrun -f notarytool) submit dist/Ollama-darwin.zip --wait --timeout 20m --apple-id "$APPLE_ID" --password "$APPLE_PASSWORD" --team-id "$APPLE_TEAM_ID"
        rm -f dist/Ollama-darwin.zip
        $(xcrun -f stapler) staple dist/Ollama.app
        ditto -c -k --keepParent dist/Ollama.app dist/Ollama-darwin.zip
@@ -206,7 +208,7 @@ _build_macapp() {
        rm -f dist/rw*.dmg
        codesign -f --timestamp -s "$APPLE_IDENTITY" --identifier ai.ollama.ollama --options=runtime dist/Ollama.dmg
-        $(xcrun -f notarytool) submit dist/Ollama.dmg --wait --timeout 10m --apple-id "$APPLE_ID" --password "$APPLE_PASSWORD" --team-id "$APPLE_TEAM_ID"
+        $(xcrun -f notarytool) submit dist/Ollama.dmg --wait --timeout 20m --apple-id "$APPLE_ID" --password "$APPLE_PASSWORD" --team-id "$APPLE_TEAM_ID"
        $(xcrun -f stapler) staple dist/Ollama.dmg
    else
        echo "WARNING: Code signing disabled, this bundle will not work for upgrade testing"
--- a/scripts/build_linux.sh
+++ b/scripts/build_linux.sh
@@ -48,53 +48,12 @@ if echo $PLATFORM | grep "amd64" > /dev/null; then
        .
 fi
 # Deduplicate CUDA libraries across mlx_* and cuda_* directories
 deduplicate_cuda_libs() {
    local base_dir="$1"
    echo "Deduplicating CUDA libraries in ${base_dir}..."
    # Find all mlx_cuda_* directories
    for mlx_dir in "${base_dir}"/lib/ollama/mlx_cuda_*; do
        [ -d "${mlx_dir}" ] || continue
        # Extract CUDA version (e.g., v12, v13)
        cuda_version=$(basename "${mlx_dir}" | sed 's/mlx_cuda_//')
        cuda_dir="${base_dir}/lib/ollama/cuda_${cuda_version}"
        # Skip if corresponding cuda_* directory doesn't exist
        [ -d "${cuda_dir}" ] || continue
        echo "  Checking ${mlx_dir} against ${cuda_dir}..."
        # Find all .so* files in mlx directory
        find "${mlx_dir}" -type f -name "*.so*" | while read mlx_file; do
            filename=$(basename "${mlx_file}")
            cuda_file="${cuda_dir}/${filename}"
            # Skip if file doesn't exist in cuda directory
            [ -f "${cuda_file}" ] || continue
            # Compare checksums
            mlx_sum=$(sha256sum "${mlx_file}" | awk '{print $1}')
            cuda_sum=$(sha256sum "${cuda_file}" | awk '{print $1}')
            if [ "${mlx_sum}" = "${cuda_sum}" ]; then
                echo "    Deduplicating ${filename}"
                # Calculate relative path from mlx_dir to cuda_dir
                rel_path="../cuda_${cuda_version}/${filename}"
                rm -f "${mlx_file}"
                ln -s "${rel_path}" "${mlx_file}"
            fi
        done
    done
 }
 # Run deduplication for each platform output directory
 if echo $PLATFORM | grep "," > /dev/null ; then
-    deduplicate_cuda_libs "./dist/linux_amd64"
+    $(dirname $0)/deduplicate_cuda_libs.sh "./dist/linux_amd64"
-    deduplicate_cuda_libs "./dist/linux_arm64"
+    $(dirname $0)/deduplicate_cuda_libs.sh "./dist/linux_arm64"
 elif echo $PLATFORM | grep "amd64\|arm64" > /dev/null ; then
-    deduplicate_cuda_libs "./dist"
+    $(dirname $0)/deduplicate_cuda_libs.sh "./dist"
 fi
 # buildx behavior changes for single vs. multiplatform
--- a/scripts/deduplicate_cuda_libs.sh
+++ b/scripts/deduplicate_cuda_libs.sh
@@ -0,0 +1,60 @@
 #!/bin/sh
 #
 # Deduplicate CUDA libraries across mlx_* and cuda_* directories
 # This script finds identical .so* files in mlx_cuda_* directories that exist
 # in corresponding cuda_* directories and replaces them with symlinks.
 #
 set -eu
 if [ $# -eq 0 ]; then
    echo "ERROR: No directory specified" >&2
    echo "Usage: $0 <base_directory>" >&2
    exit 1
 fi
 base_dir="$1"
 if [ ! -d "${base_dir}" ]; then
    echo "ERROR: Directory ${base_dir} does not exist" >&2
    exit 1
 fi
 echo "Deduplicating CUDA libraries in ${base_dir}..."
 # Find all mlx_cuda_* directories
 for mlx_dir in "${base_dir}"/lib/ollama/mlx_cuda_*; do
    [ -d "${mlx_dir}" ] || continue
    # Extract CUDA version (e.g., v12, v13)
    cuda_version=$(basename "${mlx_dir}" | sed 's/mlx_cuda_//')
    cuda_dir="${base_dir}/lib/ollama/cuda_${cuda_version}"
    # Skip if corresponding cuda_* directory doesn't exist
    [ -d "${cuda_dir}" ] || continue
    echo "  Checking ${mlx_dir} against ${cuda_dir}..."
    # Find all .so* files in mlx directory
    find "${mlx_dir}" -type f -name "*.so*" | while read mlx_file; do
        filename=$(basename "${mlx_file}")
        cuda_file="${cuda_dir}/${filename}"
        # Skip if file doesn't exist in cuda directory
        [ -f "${cuda_file}" ] || continue
        # Compare checksums
        mlx_sum=$(sha256sum "${mlx_file}" | awk '{print $1}')
        cuda_sum=$(sha256sum "${cuda_file}" | awk '{print $1}')
        if [ "${mlx_sum}" = "${cuda_sum}" ]; then
            echo "    Deduplicating ${filename}"
            # Calculate relative path from mlx_dir to cuda_dir
            rel_path="../cuda_${cuda_version}/${filename}"
            rm -f "${mlx_file}"
            ln -s "${rel_path}" "${mlx_file}"
        fi
    done
 done
 echo "Deduplication complete"
--- a/server/download.go
+++ b/server/download.go
@@ -95,11 +95,48 @@ func (p *blobDownloadPart) UnmarshalJSON(b []byte) error {
 }
 const (
 	// numDownloadParts is the default number of concurrent download parts for standard downloads
 	numDownloadParts = 16
 	// numHFDownloadParts is the reduced number of concurrent download parts for HuggingFace
 	// downloads to avoid triggering rate limits (HTTP 429 errors). See GitHub issue #13297.
 	numHFDownloadParts        = 4
 	minDownloadPartSize int64 = 100 * format.MegaByte
 	maxDownloadPartSize int64 = 1000 * format.MegaByte
 )
 // isHuggingFaceURL returns true if the URL is from a HuggingFace domain.
 // This includes:
 //   - huggingface.co (main domain)
 //   - *.huggingface.co (subdomains like cdn-lfs.huggingface.co)
 //   - hf.co (shortlink domain)
 //   - *.hf.co (CDN domains like cdn-lfs.hf.co, cdn-lfs3.hf.co)
 func isHuggingFaceURL(u *url.URL) bool {
 	if u == nil {
 		return false
 	}
 	host := strings.ToLower(u.Hostname())
 	return host == "huggingface.co" ||
 		strings.HasSuffix(host, ".huggingface.co") ||
 		host == "hf.co" ||
 		strings.HasSuffix(host, ".hf.co")
 }
 // getNumDownloadParts returns the number of concurrent download parts to use
 // for the given URL. HuggingFace URLs use reduced concurrency (default 4) to
 // avoid triggering rate limits. This can be overridden via the OLLAMA_HF_CONCURRENCY
 // environment variable. For non-HuggingFace URLs, returns the standard concurrency (16).
 func getNumDownloadParts(u *url.URL) int {
 	if isHuggingFaceURL(u) {
 		if v := os.Getenv("OLLAMA_HF_CONCURRENCY"); v != "" {
 			if n, err := strconv.Atoi(v); err == nil && n > 0 {
 				return n
 			}
 		}
 		return numHFDownloadParts
 	}
 	return numDownloadParts
 }
 func (p *blobDownloadPart) Name() string {
 	return strings.Join([]string{
 		p.blobDownload.Name, "partial", strconv.Itoa(p.N),
@@ -271,7 +308,11 @@ func (b *blobDownload) run(ctx context.Context, requestURL *url.URL, opts *regis
 	}
 	g, inner := errgroup.WithContext(ctx)
-	g.SetLimit(numDownloadParts)
+	concurrency := getNumDownloadParts(directURL)
 	if concurrency != numDownloadParts {
 		slog.Info(fmt.Sprintf("using reduced concurrency (%d) for HuggingFace download", concurrency))
 	}
 	g.SetLimit(concurrency)
 	for i := range b.Parts {
 		part := b.Parts[i]
 		if part.Completed.Load() == part.Size {
--- a/server/download_test.go
+++ b/server/download_test.go
@@ -0,0 +1,194 @@
 package server
 import (
 	"net/url"
 	"testing"
 	"github.com/stretchr/testify/assert"
 )
 func TestIsHuggingFaceURL(t *testing.T) {
 	tests := []struct {
 		name     string
 		url      string
 		expected bool
 	}{
 		{
 			name:     "nil url",
 			url:      "",
 			expected: false,
 		},
 		{
 			name:     "huggingface.co main domain",
 			url:      "https://huggingface.co/some/model",
 			expected: true,
 		},
 		{
 			name:     "cdn-lfs.huggingface.co subdomain",
 			url:      "https://cdn-lfs.huggingface.co/repos/abc/123",
 			expected: true,
 		},
 		{
 			name:     "cdn-lfs3.hf.co CDN domain",
 			url:      "https://cdn-lfs3.hf.co/repos/abc/123",
 			expected: true,
 		},
 		{
 			name:     "hf.co shortlink domain",
 			url:      "https://hf.co/model",
 			expected: true,
 		},
 		{
 			name:     "uppercase HuggingFace domain",
 			url:      "https://HUGGINGFACE.CO/model",
 			expected: true,
 		},
 		{
 			name:     "mixed case HF domain",
 			url:      "https://Cdn-Lfs.HF.Co/repos",
 			expected: true,
 		},
 		{
 			name:     "ollama registry",
 			url:      "https://registry.ollama.ai/v2/library/llama3",
 			expected: false,
 		},
 		{
 			name:     "github.com",
 			url:      "https://github.com/ollama/ollama",
 			expected: false,
 		},
 		{
 			name:     "fake huggingface domain",
 			url:      "https://nothuggingface.co/model",
 			expected: false,
 		},
 		{
 			name:     "fake hf domain",
 			url:      "https://nothf.co/model",
 			expected: false,
 		},
 		{
 			name:     "huggingface in path not host",
 			url:      "https://example.com/huggingface.co/model",
 			expected: false,
 		},
 	}
 	for _, tc := range tests {
 		t.Run(tc.name, func(t *testing.T) {
 			var u *url.URL
 			if tc.url != "" {
 				var err error
 				u, err = url.Parse(tc.url)
 				if err != nil {
 					t.Fatalf("failed to parse URL: %v", err)
 				}
 			}
 			got := isHuggingFaceURL(u)
 			assert.Equal(t, tc.expected, got)
 		})
 	}
 }
 func TestGetNumDownloadParts(t *testing.T) {
 	tests := []struct {
 		name        string
 		url         string
 		envValue    string
 		expected    int
 		description string
 	}{
 		{
 			name:        "nil url returns default",
 			url:         "",
 			envValue:    "",
 			expected:    numDownloadParts,
 			description: "nil URL should return standard concurrency",
 		},
 		{
 			name:        "ollama registry returns default",
 			url:         "https://registry.ollama.ai/v2/library/llama3",
 			envValue:    "",
 			expected:    numDownloadParts,
 			description: "Ollama registry should use standard concurrency",
 		},
 		{
 			name:        "huggingface returns reduced default",
 			url:         "https://huggingface.co/model/repo",
 			envValue:    "",
 			expected:    numHFDownloadParts,
 			description: "HuggingFace should use reduced concurrency",
 		},
 		{
 			name:        "hf.co CDN returns reduced default",
 			url:         "https://cdn-lfs3.hf.co/repos/abc/123",
 			envValue:    "",
 			expected:    numHFDownloadParts,
 			description: "HuggingFace CDN should use reduced concurrency",
 		},
 		{
 			name:        "huggingface with env override",
 			url:         "https://huggingface.co/model/repo",
 			envValue:    "2",
 			expected:    2,
 			description: "OLLAMA_HF_CONCURRENCY should override default",
 		},
 		{
 			name:        "huggingface with higher env override",
 			url:         "https://huggingface.co/model/repo",
 			envValue:    "8",
 			expected:    8,
 			description: "OLLAMA_HF_CONCURRENCY can be set higher than default",
 		},
 		{
 			name:        "huggingface with invalid env (non-numeric)",
 			url:         "https://huggingface.co/model/repo",
 			envValue:    "invalid",
 			expected:    numHFDownloadParts,
 			description: "Invalid OLLAMA_HF_CONCURRENCY should fall back to default",
 		},
 		{
 			name:        "huggingface with invalid env (zero)",
 			url:         "https://huggingface.co/model/repo",
 			envValue:    "0",
 			expected:    numHFDownloadParts,
 			description: "Zero OLLAMA_HF_CONCURRENCY should fall back to default",
 		},
 		{
 			name:        "huggingface with invalid env (negative)",
 			url:         "https://huggingface.co/model/repo",
 			envValue:    "-1",
 			expected:    numHFDownloadParts,
 			description: "Negative OLLAMA_HF_CONCURRENCY should fall back to default",
 		},
 		{
 			name:        "non-huggingface ignores env",
 			url:         "https://registry.ollama.ai/v2/library/llama3",
 			envValue:    "2",
 			expected:    numDownloadParts,
 			description: "OLLAMA_HF_CONCURRENCY should not affect non-HF URLs",
 		},
 	}
 	for _, tc := range tests {
 		t.Run(tc.name, func(t *testing.T) {
 			// Set or clear the environment variable
 			if tc.envValue != "" {
 				t.Setenv("OLLAMA_HF_CONCURRENCY", tc.envValue)
 			}
 			var u *url.URL
 			if tc.url != "" {
 				var err error
 				u, err = url.Parse(tc.url)
 				if err != nil {
 					t.Fatalf("failed to parse URL: %v", err)
 				}
 			}
 			got := getNumDownloadParts(u)
 			assert.Equal(t, tc.expected, got, tc.description)
 		})
 	}
 }
--- a/x/README.md
+++ b/x/README.md
@@ -4,21 +4,47 @@
 We're working on a new experimental backend based on the [MLX project](https://github.com/ml-explore/mlx)
-Support is currently limited to MacOS and Linux with CUDA GPUs.  We're looking to add support for Windows CUDA soon, and other GPU vendors.  To build:
+Support is currently limited to MacOS and Linux with CUDA GPUs. We're looking to add support for Windows CUDA soon, and other GPU vendors.
-```
+### Building ollama-mlx
 The `ollama-mlx` binary is a separate build of Ollama with MLX support enabled. This enables experimental features like image generation.
 #### macOS (Apple Silicon and Intel)
 ```bash
 # Build MLX backend libraries
 cmake --preset MLX
 cmake --build --preset MLX --parallel
 cmake --install build --component MLX
-go build -tags mlx .
+
 # Build ollama-mlx binary
 go build -tags mlx -o ollama-mlx .
 ```
-On linux, use the preset "MLX CUDA 13" or "MLX CUDA 12" to enable CUDA with the default Ollama NVIDIA GPU architectures enabled. 
+#### Linux (CUDA)
 On Linux, use the preset "MLX CUDA 13" or "MLX CUDA 12" to enable CUDA with the default Ollama NVIDIA GPU architectures enabled:
 ```bash
 # Build MLX backend libraries with CUDA support
 cmake --preset 'MLX CUDA 13'
 cmake --build --preset 'MLX CUDA 13' --parallel
 cmake --install build --component MLX
 # Build ollama-mlx binary
 CGO_CFLAGS="-O3 -I$(pwd)/build/_deps/mlx-c-src" \
 CGO_LDFLAGS="-L$(pwd)/build/lib/ollama -lmlxc -lmlx" \
 go build -tags mlx -o ollama-mlx .
 ```
 #### Using build scripts
 The build scripts automatically create the `ollama-mlx` binary:
 - **macOS**: `./scripts/build_darwin.sh` produces `dist/darwin/ollama-mlx`
 - **Linux**: `./scripts/build_linux.sh` produces `ollama-mlx` in the output archives
 ## Image Generation
-Based on the experimental MLX backend, we're working on adding imagegen support.  After running the cmake commands above:
+Image generation is built into the `ollama-mlx` binary. Run `ollama-mlx serve` to start the server with image generation support enabled.
 ```
 go build -o imagegen ./x/imagegen/cmd/engine
 ```
--- a/x/imagegen/cli.go
+++ b/x/imagegen/cli.go
@@ -123,11 +123,6 @@ func RegisterFlags(cmd *cobra.Command) {
 // Returns true if it handled the request, false if the caller should continue with normal flow.
 // Supports flags: --width, --height, --steps, --seed, --negative
 func RunCLI(cmd *cobra.Command, name string, prompt string, interactive bool, keepAlive *api.Duration) error {
 	// Verify it's a valid image gen model
 	if ResolveModelName(name) == "" {
 		return fmt.Errorf("unknown image generation model: %s", name)
 	}
 	// Get options from flags (with env var defaults)
 	opts := DefaultOptions()
 	if cmd != nil && cmd.Flags() != nil {
@@ -511,10 +506,7 @@ func displayImageInTerminal(imagePath string) bool {
 		// Send in chunks for large images
 		const chunkSize = 4096
 		for i := 0; i < len(encoded); i += chunkSize {
-			end := i + chunkSize
+			end := min(i+chunkSize, len(encoded))
 			if end > len(encoded) {
 				end = len(encoded)
 			}
 			chunk := encoded[i:end]
 			if i == 0 {
--- a/x/imagegen/server.go
+++ b/x/imagegen/server.go
@@ -14,7 +14,9 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
 	"runtime"
 	"strconv"
 	"strings"
 	"sync"
 	"time"
@@ -70,7 +72,7 @@ func NewServer(modelName string) (*Server, error) {
 		port = rand.Intn(65535-49152) + 49152
 	}
-	// Get the ollama executable path
+	// Get the ollama-mlx executable path (in same directory as current executable)
 	exe, err := os.Executable()
 	if err != nil {
 		return nil, fmt.Errorf("unable to lookup executable path: %w", err)
@@ -78,11 +80,42 @@ func NewServer(modelName string) (*Server, error) {
 	if eval, err := filepath.EvalSymlinks(exe); err == nil {
 		exe = eval
 	}
 	mlxExe := filepath.Join(filepath.Dir(exe), "ollama-mlx")
-	// Spawn subprocess: ollama runner --image-engine --model <path> --port <port>
+	// Spawn subprocess: ollama-mlx runner --image-engine --model <path> --port <port>
-	cmd := exec.Command(exe, "runner", "--image-engine", "--model", modelName, "--port", strconv.Itoa(port))
+	cmd := exec.Command(mlxExe, "runner", "--image-engine", "--model", modelName, "--port", strconv.Itoa(port))
 	cmd.Env = os.Environ()
 	// On Linux, set LD_LIBRARY_PATH to include MLX library directories
 	if runtime.GOOS == "linux" {
 		// Build library paths: start with LibOllamaPath, then add any mlx_* subdirectories
 		libraryPaths := []string{ml.LibOllamaPath}
 		if mlxDirs, err := filepath.Glob(filepath.Join(ml.LibOllamaPath, "mlx_*")); err == nil {
 			libraryPaths = append(libraryPaths, mlxDirs...)
 		}
 		// Append existing LD_LIBRARY_PATH if set
 		if existingPath, ok := os.LookupEnv("LD_LIBRARY_PATH"); ok {
 			libraryPaths = append(libraryPaths, filepath.SplitList(existingPath)...)
 		}
 		pathEnvVal := strings.Join(libraryPaths, string(filepath.ListSeparator))
 		// Update or add LD_LIBRARY_PATH in cmd.Env
 		found := false
 		for i := range cmd.Env {
 			if strings.HasPrefix(cmd.Env[i], "LD_LIBRARY_PATH=") {
 				cmd.Env[i] = "LD_LIBRARY_PATH=" + pathEnvVal
 				found = true
 				break
 			}
 		}
 		if !found {
 			cmd.Env = append(cmd.Env, "LD_LIBRARY_PATH="+pathEnvVal)
 		}
 		slog.Debug("mlx subprocess library path", "LD_LIBRARY_PATH", pathEnvVal)
 	}
 	s := &Server{
 		cmd:       cmd,
 		port:      port,
@@ -113,7 +146,7 @@ func NewServer(modelName string) (*Server, error) {
 		}
 	}()
-	slog.Info("starting image runner subprocess", "model", modelName, "port", port)
+	slog.Info("starting ollama-mlx image runner subprocess", "exe", mlxExe, "model", modelName, "port", port)
 	if err := cmd.Start(); err != nil {
 		return nil, fmt.Errorf("failed to start image runner: %w", err)
 	}
Author	SHA1	Message	Date
Parth Sareen	6b2abfb433	server: add tests and fix isHuggingFaceURL edge case - Add comprehensive tests for isHuggingFaceURL and getNumDownloadParts - Fix bug where domains ending in huggingface.co (like nothuggingface.co) would incorrectly match as HuggingFace URLs - Improve code comments with more detailed documentation	2026-01-18 16:45:17 -08:00
Parth Sareen	805ed4644c	server: reduce download concurrency for HuggingFace URLs Reduces concurrent download parts from 16 to 4 for HuggingFace URLs to avoid triggering rate limits (HTTP 429 errors). Adds OLLAMA_HF_CONCURRENCY environment variable for users who want to customize the concurrency level. Fixes #13297	2026-01-18 16:38:49 -08:00
Daniel Hiltgen	e4b488a7b5	CI: dedup cuda libraries to reduce payload size (#13704 )	2026-01-13 11:25:31 -08:00
Daniel Hiltgen	98079ddd79	ci: add missing mlx components to release build (#13702 )	2026-01-13 09:13:09 -08:00
Jeffrey Morgan	d70942f47b	x/imagegen/cli: skip local model check (#13699 )	2026-01-12 22:38:10 -08:00
Jeffrey Morgan	58e4701557	scripts: increase notarization timeout to 20m (#13697 ) The 100MB mlx.metallib file significantly increased the app bundle size, causing Apple's notarization service to timeout with the previous 10m limit.	2026-01-12 20:38:38 -08:00
Jeffrey Morgan	dbf47ee55a	cmake: use CMAKE_SYSTEM_PROCESSOR instead of CMAKE_OSX_ARCHITECTURES for mlx.metallib install (#13696 ) The CMake condition for installing mlx.metallib checks CMAKE_OSX_ARCHITECTURES, but this variable is only set when explicitly passed - not auto-detected. The arm64 build was missing this flag, causing the metallib to not be installed, which then caused codesign to fail on the unexpanded glob pattern.	2026-01-12 20:05:11 -08:00
Jeffrey Morgan	af7ea6e96e	x/imagegen: install mlx.metallib and fix macOS rpath handling, add mlx library directories to LD_LIBRARY_PATH (#13695 ) - Install mlx.metallib for arm64 builds (required for Metal GPU acceleration) - Apply rpath settings to all macOS builds, not just x86_64 - Add CMAKE_BUILD_WITH_INSTALL_RPATH to avoid install_name_tool errors - Update build_darwin.sh to copy, sign, and package the metallib	2026-01-12 19:03:11 -08:00
Jeffrey Morgan	8f1e0140e7	x/imagegen: fix mlx build in Dockerfile and macOS build script (#13693 )	2026-01-12 15:52:43 -08:00