Add managed local llama.cpp embeddings

README.md

@@ -50,13 +50,23 @@ curl -sSL https://raw.githubusercontent.com/yoanbernabeu/grepai/main/install.sh
 irm https://raw.githubusercontent.com/yoanbernabeu/grepai/main/install.ps1 | iex
 ```
 
-Requires an embedding provider — [Ollama](https://ollama.ai) (default), [LM Studio](https://lmstudio.ai), or OpenAI.
+Requires an embedding provider — [Ollama](https://ollama.ai) (default), managed local `llama.cpp`, [LM Studio](https://lmstudio.ai), or OpenAI.
 
 **Ollama (recommended):**
 ```bash
 ollama pull nomic-embed-text
 ```
 
+**Managed local `llama.cpp`:**
+```bash
+grepai init --provider llamacpp
+grepai model install
+grepai model use bge-small-en-v1.5-q8_0
+```
+
+If you already have managed local models installed, plain `grepai init` will ask which installed `llamacpp` model to use when you choose the `llamacpp` provider.
+Managed `llama.cpp` runtime support is currently limited to macOS (`arm64`, `amd64`), Linux (`amd64`), and Windows (`amd64`).
+
 ## Quick Start
 
 ```bash
@@ -68,7 +78,7 @@ grepai trace callers "Login"       # Find who calls a function
 
 ## Shell Completion
 
-grepai supports autocompletion for commands, flags, and dynamic values (workspace names, project names, providers, backends).
+grepai supports autocompletion for commands, flags, and dynamic values (workspace names, project names, providers, backends, and managed model ids for `llamacpp`).
 
 **Zsh (add to `~/.zshrc`):**
 ```bash

cli/completion.go

@@ -3,6 +3,7 @@ package cli
 import (
 	"github.com/spf13/cobra"
 	"github.com/yoanbernabeu/grepai/config"
+	"github.com/yoanbernabeu/grepai/internal/managedassets"
 )
 
 var completionCmd = &cobra.Command{
@@ -92,13 +93,17 @@ func init() {
 func registerCompletions() {
 	// Static flag completions for initCmd
 	_ = initCmd.RegisterFlagCompletionFunc("provider", func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
-		return []string{
+		completions := []string{
 			"ollama\tLocal embedding with Ollama",
 			"lmstudio\tLocal embedding with LM Studio",
 			"openai\tCloud embedding with OpenAI",
 			"synthetic\tCloud embedding with Synthetic (free)",
 			"openrouter\tCloud multi-provider gateway",
-		}, cobra.ShellCompDirectiveNoFileComp
+		}
+		if managedLlamaCPPSupported() {
+			completions = append(completions, "llamacpp\tManaged local embedding with llama.cpp")
+		}
+		return completions, cobra.ShellCompDirectiveNoFileComp
 	})
 	_ = initCmd.RegisterFlagCompletionFunc("backend", func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
 		return []string{
@@ -107,6 +112,26 @@ func registerCompletions() {
 			"qdrant\tQdrant vector database",
 		}, cobra.ShellCompDirectiveNoFileComp
 	})
+	_ = initCmd.RegisterFlagCompletionFunc("model", func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
+		provider, _ := cmd.Flags().GetString("provider")
+		switch provider {
+		case "llamacpp":
+			return completeAvailableManagedModels(), cobra.ShellCompDirectiveNoFileComp
+		case "openai":
+			return []string{
+				"text-embedding-3-small\tOpenAI small embedding model",
+				"text-embedding-3-large\tOpenAI large embedding model",
+			}, cobra.ShellCompDirectiveNoFileComp
+		case "openrouter":
+			return []string{
+				"openai/text-embedding-3-small\tOpenRouter small embedding model",
+				"openai/text-embedding-3-large\tOpenRouter large embedding model",
+				"qwen/qwen3-embedding-8b\tOpenRouter Qwen code-focused embedding model",
+			}, cobra.ShellCompDirectiveNoFileComp
+		default:
+			return nil, cobra.ShellCompDirectiveNoFileComp
+		}
+	})
 
 	// Static flag completions for workspaceCreateCmd
 	_ = workspaceCreateCmd.RegisterFlagCompletionFunc("backend", func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
@@ -116,15 +141,38 @@ func registerCompletions() {
 		}, cobra.ShellCompDirectiveNoFileComp
 	})
 	_ = workspaceCreateCmd.RegisterFlagCompletionFunc("provider", func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
-		return []string{
+		completions := []string{
 			"ollama\tLocal embedding with Ollama",
 			"lmstudio\tLocal embedding with LM Studio",
 			"openai\tCloud embedding with OpenAI",
 			"synthetic\tCloud embedding with Synthetic (free)",
 			"openrouter\tCloud multi-provider gateway",
-		}, cobra.ShellCompDirectiveNoFileComp
+		}
+		if managedLlamaCPPSupported() {
+			completions = append(completions, "llamacpp\tManaged local embedding with llama.cpp")
+		}
+		return completions, cobra.ShellCompDirectiveNoFileComp
 	})
 
+	modelUseCmd.ValidArgsFunction = func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
+		if len(args) == 0 {
+			return completeInstalledManagedModels(), cobra.ShellCompDirectiveNoFileComp
+		}
+		return nil, cobra.ShellCompDirectiveNoFileComp
+	}
+	modelRemoveCmd.ValidArgsFunction = func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
+		if len(args) == 0 {
+			return completeInstalledManagedModels(), cobra.ShellCompDirectiveNoFileComp
+		}
+		return nil, cobra.ShellCompDirectiveNoFileComp
+	}
+	modelInstallCmd.ValidArgsFunction = func(cmd *cobra.Command, args []string, toComplete string) ([]string, cobra.ShellCompDirective) {
+		if len(args) == 0 {
+			return completeAvailableManagedModels(), cobra.ShellCompDirectiveNoFileComp
+		}
+		return nil, cobra.ShellCompDirectiveNoFileComp
+	}
+
 	// Static flag completions for trace commands (mode)
 	for _, cmd := range []*cobra.Command{traceCallersCmd, traceCalleesCmd, traceGraphCmd} {
 		cmd := cmd
@@ -218,3 +266,24 @@ func completeProjectNames(workspaceName string) []string {
 	}
 	return names
 }
+
+func completeAvailableManagedModels() []string {
+	models := managedassets.ListAvailableModels()
+	completions := make([]string, 0, len(models))
+	for _, model := range models {
+		completions = append(completions, model.ID+"\t"+model.Display)
+	}
+	return completions
+}
+
+func completeInstalledManagedModels() []string {
+	models, err := managedassets.LoadInstalledModels()
+	if err != nil {
+		return nil
+	}
+	completions := make([]string, 0, len(models))
+	for _, model := range models {
+		completions = append(completions, model.ID+"\tinstalled managed model")
+	}
+	return completions
+}

cli/completion_test.go

@@ -8,9 +8,17 @@ import (
 	"testing"
 
 	"github.com/yoanbernabeu/grepai/config"
+	"github.com/yoanbernabeu/grepai/internal/managedassets"
 )
 
 func TestCompletionZsh_should_output_compdef(t *testing.T) {
+	prevProvider := initProvider
+	prevModel := initModel
+	defer func() {
+		initProvider = prevProvider
+		initModel = prevModel
+	}()
+
 	var buf bytes.Buffer
 	rootCmd.SetOut(&buf)
 	rootCmd.SetArgs([]string{"completion", "zsh"})
@@ -30,6 +38,13 @@ func TestCompletionZsh_should_output_compdef(t *testing.T) {
 }
 
 func TestCompletionBash_should_output_valid_script(t *testing.T) {
+	prevProvider := initProvider
+	prevModel := initModel
+	defer func() {
+		initProvider = prevProvider
+		initModel = prevModel
+	}()
+
 	var buf bytes.Buffer
 	rootCmd.SetOut(&buf)
 	rootCmd.SetArgs([]string{"completion", "bash"})
@@ -45,6 +60,13 @@ func TestCompletionBash_should_output_valid_script(t *testing.T) {
 }
 
 func TestCompletionFish_should_output_valid_script(t *testing.T) {
+	prevProvider := initProvider
+	prevModel := initModel
+	defer func() {
+		initProvider = prevProvider
+		initModel = prevModel
+	}()
+
 	var buf bytes.Buffer
 	rootCmd.SetOut(&buf)
 	rootCmd.SetArgs([]string{"completion", "fish"})
@@ -60,6 +82,13 @@ func TestCompletionFish_should_output_valid_script(t *testing.T) {
 }
 
 func TestCompletionPowershell_should_output_valid_script(t *testing.T) {
+	prevProvider := initProvider
+	prevModel := initModel
+	defer func() {
+		initProvider = prevProvider
+		initModel = prevModel
+	}()
+
 	var buf bytes.Buffer
 	rootCmd.SetOut(&buf)
 	rootCmd.SetArgs([]string{"completion", "powershell"})
@@ -133,3 +162,93 @@ func TestCompleteProjectNames_should_return_project_names(t *testing.T) {
 		t.Fatalf("expected frontend and backend, got: %v", names)
 	}
 }
+
+func TestCompletionScriptIncludesLlamaCPPProvider(t *testing.T) {
+	prevProvider := initProvider
+	prevModel := initModel
+	defer func() {
+		initProvider = prevProvider
+		initModel = prevModel
+	}()
+
+	var buf bytes.Buffer
+	rootCmd.SetOut(&buf)
+	rootCmd.SetArgs([]string{"__complete", "init", "--provider", ""})
+	defer rootCmd.SetOut(nil)
+
+	if err := rootCmd.Execute(); err != nil {
+		t.Fatalf("provider completion failed: %v", err)
+	}
+
+	if !strings.Contains(buf.String(), "llamacpp") {
+		t.Fatalf("expected llamacpp in completion output, got: %s", buf.String())
+	}
+}
+
+func TestCompletionInitModelIncludesManagedModelsForLlamaCPP(t *testing.T) {
+	prevProvider := initProvider
+	prevModel := initModel
+	defer func() {
+		initProvider = prevProvider
+		initModel = prevModel
+	}()
+
+	var buf bytes.Buffer
+	rootCmd.SetOut(&buf)
+	rootCmd.SetArgs([]string{"__complete", "init", "--provider", "llamacpp", "--model", ""})
+	defer rootCmd.SetOut(nil)
+
+	if err := rootCmd.Execute(); err != nil {
+		t.Fatalf("model completion failed: %v", err)
+	}
+
+	output := buf.String()
+	if !strings.Contains(output, "bge-small-en-v1.5-q8_0") {
+		t.Fatalf("expected default managed model in completion output, got: %s", output)
+	}
+	if !strings.Contains(output, "nomic-embed-text-v1.5-q8_0") {
+		t.Fatalf("expected Nomic managed model in completion output, got: %s", output)
+	}
+}
+
+func TestCompletionModelUseIncludesInstalledModels(t *testing.T) {
+	prevProvider := initProvider
+	prevModel := initModel
+	defer func() {
+		initProvider = prevProvider
+		initModel = prevModel
+	}()
+
+	tmpDir := t.TempDir()
+	oldHome := os.Getenv("HOME")
+	_ = os.Setenv("HOME", tmpDir)
+	defer os.Setenv("HOME", oldHome)
+
+	modelDef, err := managedassets.LookupModel("nomic-embed-text-v1.5-q8_0")
+	if err != nil {
+		t.Fatalf("LookupModel failed: %v", err)
+	}
+	if err := managedassets.SaveInstalledModels([]managedassets.InstalledModel{{
+		ID:         modelDef.ID,
+		FileName:   modelDef.FileName,
+		Path:       filepath.Join(tmpDir, modelDef.FileName),
+		SourceURL:  modelDef.URL,
+		SizeBytes:  modelDef.SizeBytes,
+		Dimensions: modelDef.Dimensions,
+	}}); err != nil {
+		t.Fatalf("SaveInstalledModels failed: %v", err)
+	}
+
+	var buf bytes.Buffer
+	rootCmd.SetOut(&buf)
+	rootCmd.SetArgs([]string{"__complete", "model", "use", ""})
+	defer rootCmd.SetOut(nil)
+
+	if err := rootCmd.Execute(); err != nil {
+		t.Fatalf("model use completion failed: %v", err)
+	}
+
+	if !strings.Contains(buf.String(), modelDef.ID) {
+		t.Fatalf("expected installed model in completion output, got: %s", buf.String())
+	}
+}