feat: Voyage ai provider

CHANGELOG.md

@@ -9,6 +9,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
+- **Voyage AI Provider**: Add Voyage AI as a new cloud embedding provider (`voyageai`) with code-optimized models like `voyage-code-3` (1024 dims), batch embedding, adaptive rate limiting, and full integration across CLI, MCP server, and workspaces
 - **`.grepaiignore` Support**: New `.grepaiignore` file allows overriding `.gitignore` rules for grepai indexing. Supports negation patterns (`!`) to re-include files excluded by `.gitignore`, with directory-level precedence for nested files (#107)
 
 ## [0.34.0] - 2026-02-24

README.md

@@ -50,7 +50,7 @@ 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), [LM Studio](https://lmstudio.ai), OpenAI, or [Voyage AI](https://voyageai.com).
 
 **Ollama (recommended):**
 ```bash

cli/init.go

@@ -32,14 +32,14 @@ var initCmd = &cobra.Command{
 
 This command will:
 - Create .grepai/config.yaml with default settings
-- Prompt for embedding provider (Ollama or OpenAI)
+- Prompt for embedding provider (Ollama, LM Studio, OpenAI, or Voyage AI)
 - Prompt for storage backend (GOB file or PostgreSQL)
 - Add .grepai/ to .gitignore if present`,
 	RunE: runInit,
 }
 
 func init() {
-	initCmd.Flags().StringVarP(&initProvider, "provider", "p", "", "Embedding provider (ollama, lmstudio, openai, synthetic, or openrouter)")
+	initCmd.Flags().StringVarP(&initProvider, "provider", "p", "", "Embedding provider (ollama, lmstudio, openai, voyageai, synthetic, or openrouter)")
 	initCmd.Flags().StringVarP(&initModel, "model", "m", "", "Embedding model (for openrouter: text-embedding-3-small, text-embedding-3-large, qwen3-embedding-8b)")
 	initCmd.Flags().StringVarP(&initBackend, "backend", "b", "", "Storage backend (gob, postgres, or qdrant)")
 	initCmd.Flags().BoolVar(&initNonInteractive, "yes", false, "Use defaults without prompting")
@@ -123,6 +123,7 @@ func runInit(cmd *cobra.Command, args []string) error {
 			fmt.Println("  3) openai (cloud, requires API key)")
 			fmt.Println("  4) synthetic (cloud, free embedding API)")
 			fmt.Println("  5) openrouter (cloud, multi-provider gateway)")
+			fmt.Println("  6) voyageai (cloud, optimized for code, requires API key)")
 			fmt.Print("Choice [1]: ")
 
 			input, _ := reader.ReadString('\n')
@@ -145,7 +146,7 @@ func runInit(cmd *cobra.Command, args []string) error {
 				cfg.Embedder.Provider = "openai"
 				cfg.Embedder.Model = "text-embedding-3-small"
 				cfg.Embedder.Endpoint = "https://api.openai.com/v1"
-				// OpenAI: leave Dimensions nil to use model's native dimensions
+				cfg.Embedder.Dimensions = nil // use model's native dimensions
 			case "4", "synthetic":
 				cfg.Embedder.Provider = "synthetic"
 				cfg.Embedder.Model = "hf:nomic-ai/nomic-embed-text-v1.5"
@@ -155,7 +156,7 @@ func runInit(cmd *cobra.Command, args []string) error {
 			case "5", "openrouter":
 				cfg.Embedder.Provider = "openrouter"
 				cfg.Embedder.Endpoint = "https://openrouter.ai/api/v1"
-				// OpenRouter: leave Dimensions nil to use model's native dimensions
+				cfg.Embedder.Dimensions = nil // use model's native dimensions
 
 				// Model selection for OpenRouter
 				fmt.Println("\nSelect OpenRouter embedding model:")
@@ -175,6 +176,11 @@ func runInit(cmd *cobra.Command, args []string) error {
 				default:
 					cfg.Embedder.Model = "openai/text-embedding-3-small"
 				}
+			case "6", "voyageai":
+				cfg.Embedder.Provider = "voyageai"
+				cfg.Embedder.Model = "voyage-code-3"
+				cfg.Embedder.Endpoint = "https://api.voyageai.com/v1"
+				cfg.Embedder.Dimensions = nil // use model's native dimensions (1024)
 			default:
 				cfg.Embedder.Provider = "ollama"
 				fmt.Print("Ollama endpoint [http://localhost:11434]: ")
@@ -196,7 +202,11 @@ func runInit(cmd *cobra.Command, args []string) error {
 			case "openai":
 				cfg.Embedder.Model = "text-embedding-3-small"
 				cfg.Embedder.Endpoint = "https://api.openai.com/v1"
-				// OpenAI: leave Dimensions nil to use model's native dimensions
+				cfg.Embedder.Dimensions = nil // use model's native dimensions
+			case "voyageai":
+				cfg.Embedder.Model = "voyage-code-3"
+				cfg.Embedder.Endpoint = "https://api.voyageai.com/v1"
+				cfg.Embedder.Dimensions = nil // use model's native dimensions (1024)
 			case "synthetic":
 				cfg.Embedder.Model = "hf:nomic-ai/nomic-embed-text-v1.5"
 				cfg.Embedder.Endpoint = "https://api.synthetic.new/openai/v1"
@@ -205,7 +215,7 @@ func runInit(cmd *cobra.Command, args []string) error {
 			case "openrouter":
 				cfg.Embedder.Model = "openai/text-embedding-3-small"
 				cfg.Embedder.Endpoint = "https://openrouter.ai/api/v1"
-				// OpenRouter: leave Dimensions nil to use model's native dimensions
+				cfg.Embedder.Dimensions = nil // use model's native dimensions
 			}
 		}
 
@@ -339,6 +349,8 @@ func runInit(cmd *cobra.Command, args []string) error {
 		fmt.Printf("  Endpoint: %s\n", cfg.Embedder.Endpoint)
 	case "openai":
 		fmt.Println("\nMake sure OPENAI_API_KEY is set in your environment.")
+	case "voyageai":
+		fmt.Println("\nMake sure VOYAGE_API_KEY is set in your environment.")
 	case "synthetic":
 		fmt.Println("\nMake sure SYNTHETIC_API_KEY or OPENAI_API_KEY is set in your environment.")
 		fmt.Println("  Get your free API key at: https://api.synthetic.new")

cli/watch.go

@@ -424,6 +424,7 @@ func initializeEmbedder(ctx context.Context, cfg *config.Config) (embedder.Embed
 				return nil, fmt.Errorf("cannot connect to Ollama: %w\nMake sure Ollama is running and has the %s model", err, cfg.Embedder.Model)
 			}
 		}
+
 	case "lmstudio":
 		if p, ok := emb.(pinger); ok {
 			if err := p.Ping(ctx); err != nil {

config/config.go

@@ -24,10 +24,12 @@ const (
 	DefaultOpenAIEmbeddingModel     = "text-embedding-3-small"
 	DefaultSyntheticEmbeddingModel  = "hf:nomic-ai/nomic-embed-text-v1.5"
 	DefaultOpenRouterEmbeddingModel = "openai/text-embedding-3-small"
+	DefaultVoyageAIEmbeddingModel   = "voyage-code-3"
 
 	DefaultOllamaEndpoint     = "http://localhost:11434"
 	DefaultLMStudioEndpoint   = "http://127.0.0.1:1234"
 	DefaultOpenAIEndpoint     = "https://api.openai.com/v1"
+	DefaultVoyageAIEndpoint   = "https://api.voyageai.com/v1"
 	DefaultSyntheticEndpoint  = "https://api.synthetic.new/openai/v1"
 	DefaultOpenRouterEndpoint = "https://openrouter.ai/api/v1"
 
@@ -93,16 +95,17 @@ type BoostRule struct {
 }
 
 type EmbedderConfig struct {
-	Provider    string `yaml:"provider"` // ollama | lmstudio | openai | synthetic | openrouter
+	Provider    string `yaml:"provider"` // ollama | lmstudio | openai | voyageai | synthetic | openrouter
 	Model       string `yaml:"model"`
 	Endpoint    string `yaml:"endpoint,omitempty"`
 	APIKey      string `yaml:"api_key,omitempty"`
 	Dimensions  *int   `yaml:"dimensions,omitempty"`
-	Parallelism int    `yaml:"parallelism"` // Number of parallel workers for batch embedding (default: 4)
+	Parallelism int    `yaml:"parallelism,omitempty"` // Number of parallel workers for batch embedding (default: 4)
 }
 
 // GetDimensions returns the configured dimensions or a default value.
 // For OpenAI/OpenRouter, defaults to 1536 (text-embedding-3-small).
+// For Voyage AI, defaults to 1024 (voyage-code-3).
 // For Ollama/LMStudio/Synthetic, defaults to 768 (nomic-embed-text-v1.5).
 func (e *EmbedderConfig) GetDimensions() int {
 	if e.Dimensions != nil {
@@ -111,13 +114,22 @@ func (e *EmbedderConfig) GetDimensions() int {
 	switch e.Provider {
 	case "openai", "openrouter":
 		return DefaultOpenAIDimensions
+	case "voyageai":
+		return 1024
 	default:
 		return DefaultLocalEmbeddingDimensions
 	}
 }
 
 func DefaultEmbedderForProvider(provider string) EmbedderConfig {
 	switch provider {
+	case "voyageai":
+		return EmbedderConfig{
+			Provider:   "voyageai",
+			Model:      DefaultVoyageAIEmbeddingModel,
+			Endpoint:   DefaultVoyageAIEndpoint,
+			Dimensions: nil, // Voyage AI uses native dimensions (1024)
+		}
 	case "synthetic":
 		dim := DefaultLocalEmbeddingDimensions
 		return EmbedderConfig{
@@ -448,7 +460,7 @@ func (c *Config) applyDefaults() {
 		}
 	}
 
-	// Parallelism default (only used by OpenAI embedder)
+	// Parallelism default (used by OpenAI and Voyage AI embedders)
 	if c.Embedder.Parallelism <= 0 {
 		c.Embedder.Parallelism = 4
 	}

config/config_test.go

@@ -83,6 +83,14 @@ func TestDefaultEmbedderForProvider(t *testing.T) {
 	if openai.Dimensions != nil {
 		t.Fatalf("openai dimensions should be nil, got %v", openai.Dimensions)
 	}
+
+	voyageai := DefaultEmbedderForProvider("voyageai")
+	if voyageai.Endpoint != DefaultVoyageAIEndpoint || voyageai.Model != DefaultVoyageAIEmbeddingModel {
+		t.Fatalf("unexpected voyageai defaults: %+v", voyageai)
+	}
+	if voyageai.Dimensions != nil {
+		t.Fatalf("voyageai dimensions should be nil, got %v", voyageai.Dimensions)
+	}
 }
 
 func TestDefaultStoreForBackend(t *testing.T) {

docs/src/content/docs/backends/embedders.md

@@ -12,6 +12,7 @@ Embedders convert text (code chunks) into vector representations that enable sem
 | Ollama | Local | Privacy, free, no internet | Requires local resources |
 | LM Studio | Local | Privacy, OpenAI-compatible API, GUI | Requires local resources |
 | OpenAI | Cloud | High quality, fast | Costs money, sends code to cloud |
+| Voyage AI | Cloud | Optimized for code, high quality | Costs money, sends code to cloud |
 
 ## Ollama (Local)
 
@@ -233,6 +234,59 @@ For a typical codebase:
 - Initial index: ~$0.001 with `text-embedding-3-small`
 - Ongoing updates: negligible
 
+## Voyage AI (Cloud)
+
+[Voyage AI](https://voyageai.com) provides embedding models specifically optimized for code search and retrieval.
+
+### Setup
+
+1. Get an API key from [Voyage AI Dashboard](https://dash.voyageai.com/api-keys)
+
+2. Set the environment variable:
+
+```bash
+export VOYAGE_API_KEY=pa-...
+```
+
+### Configuration
+
+```yaml
+embedder:
+  provider: voyageai
+  model: voyage-code-3
+  endpoint: https://api.voyageai.com/v1
+  api_key: ${VOYAGE_API_KEY}
+```
+
+### Available Models
+
+| Model | Dimensions | Context | Notes |
+|-------|------------|---------|-------|
+| `voyage-code-3` | 1024 | 32K | Optimized for code retrieval (recommended) |
+| `voyage-4-large` | 1024 | 32K | Best general-purpose retrieval quality |
+| `voyage-4` | 1024 | 32K | Balanced quality and performance |
+| `voyage-4-lite` | 1024 | 32K | Optimized for latency and cost |
+
+All Voyage 4 series models support flexible dimensions (256, 512, 1024, 2048) and share a compatible embedding space.
+
+### Parallelism & Rate Limiting
+
+Voyage AI embeddings support parallel batch processing with adaptive rate limiting:
+
+```yaml
+embedder:
+  provider: voyageai
+  model: voyage-code-3
+  api_key: ${VOYAGE_API_KEY}
+  parallelism: 4  # Concurrent API requests (default: 4)
+```
+
+Rate limiting works the same as OpenAI: on 429 responses, parallelism auto-reduces and retries with exponential backoff.
+
+### Cost Estimation
+
+See [Voyage AI Pricing](https://docs.voyageai.com/docs/pricing) for current rates.
+
 ## Changing Embedding Models
 
 You can use any embedding model available on your provider. Two parameters matter:

docs/src/content/docs/commands/grepai_init.md

@@ -13,7 +13,7 @@ Initialize grepai by creating a .grepai directory with configuration.
 
 This command will:
 - Create .grepai/config.yaml with default settings
-- Prompt for embedding provider (Ollama or OpenAI)
+- Prompt for embedding provider (Ollama, OpenAI, Voyage AI, etc.)
 - Prompt for storage backend (GOB file or PostgreSQL)
 - Add .grepai/ to .gitignore if present
 
@@ -27,7 +27,7 @@ grepai init [flags]
   -b, --backend string    Storage backend (gob, postgres, or qdrant)
   -h, --help              help for init
       --inherit           Inherit configuration from main worktree (for git worktrees)
-  -p, --provider string   Embedding provider (ollama, lmstudio, or openai)
+  -p, --provider string   Embedding provider (ollama, lmstudio, openai, voyageai, synthetic, or openrouter)
       --ui                Run interactive Bubble Tea UI wizard
       --yes               Use defaults without prompting
 ```

docs/src/content/docs/configuration.md

@@ -17,7 +17,7 @@ version: 1
 
 # Embedder configuration
 embedder:
-  # Provider: "ollama" (local), "lmstudio" (local), or "openai" (cloud)
+  # Provider: "ollama" (local), "lmstudio" (local), "openai" (cloud), or "voyageai" (cloud)
   provider: ollama
   # Model name (depends on provider)
   model: nomic-embed-text

docs/src/content/docs/contributing.md

@@ -86,8 +86,10 @@ grepai/
 ├── config/           # Configuration loading
 ├── embedder/         # Embedding providers
 │   ├── embedder.go   # Interface
+│   ├── factory.go    # Provider factory
 │   ├── ollama.go
-│   └── openai.go
+│   ├── openai.go
+│   └── voyageai.go
 ├── store/            # Vector storage
 │   ├── store.go      # Interface
 │   ├── gob.go

docs/src/content/docs/installation.md

@@ -5,7 +5,7 @@ description: How to install grepai
 
 ## Prerequisites
 
-- **Ollama** (for local embeddings) or an **OpenAI API key** (for cloud embeddings)
+- **Ollama** (for local embeddings) or a cloud API key (**OpenAI**, **Voyage AI**, etc.)
 
 ## Homebrew (macOS)
 

docs/src/content/docs/skills.md

@@ -105,6 +105,7 @@ Install skills by category:
 | `grepai-embeddings-ollama` | Configure Ollama for local, private embeddings |
 | `grepai-embeddings-openai` | Configure OpenAI for cloud embeddings |
 | `grepai-embeddings-lmstudio` | Configure LM Studio with GUI interface |
+| `grepai-embeddings-voyageai` | Configure Voyage AI for code-optimized cloud embeddings |
 
 ### Storage Backends
 | Skill | Description |

docs/src/content/docs/workspace.md

@@ -53,7 +53,7 @@ grepai workspace create my-fullstack --from workspace-config.yaml
 | Flag | Description | Default |
 |------|-------------|---------|
 | `--backend` | Storage backend (`qdrant` or `postgres`) | Required (or `--yes`) |
-| `--provider` | Embedding provider (`ollama`, `openai`, `lmstudio`) | `ollama` with `--yes` |
+| `--provider` | Embedding provider (`ollama`, `openai`, `lmstudio`, `voyageai`, `synthetic`, `openrouter`) | `ollama` with `--yes` |
 | `--model` | Embedding model name | Provider default |
 | `--endpoint` | Embedder endpoint URL | Provider default |
 | `--dsn` | PostgreSQL connection string | Required for postgres |