Voyage AI Embedding Provider

The Voyage AI embedding provider implements the embedding.Embedder interface using the Voyage Embed API. Voyage models are optimized for retrieval tasks and support input type differentiation for asymmetric search.

Choose Voyage embeddings when retrieval quality is critical, especially for code search (voyage-code-2) or when you need asymmetric document/query embeddings. Voyage models consistently rank highly on retrieval benchmarks and offer specialized models for code, legal, and financial domains.

Installation

go get github.com/lookatitude/beluga-ai/rag/embedding/providers/voyage

Quick Start

package main

import (
    "context"
    "fmt"
    "log"
    "os"

    "github.com/lookatitude/beluga-ai/config"
    "github.com/lookatitude/beluga-ai/rag/embedding"
    _ "github.com/lookatitude/beluga-ai/rag/embedding/providers/voyage"
)

func main() {
    emb, err := embedding.New("voyage", config.ProviderConfig{
        APIKey: os.Getenv("VOYAGE_API_KEY"),
    })
    if err != nil {
        log.Fatal(err)
    }

    ctx := context.Background()
    vec, err := emb.EmbedSingle(ctx, "Beluga AI is a Go framework for agentic systems")
    if err != nil {
        log.Fatal(err)
    }

    fmt.Printf("Vector length: %d\n", len(vec))
    fmt.Printf("Dimensions: %d\n", emb.Dimensions())
}

Configuration

Parameter	Type	Default	Description
APIKey	string	(required)	Voyage AI API key
Model	string	voyage-2	Embedding model name
BaseURL	string	https://api.voyageai.com/v1	API base URL
Timeout	time.Duration	0 (no timeout)	Request timeout
Options["input_type"]	string	document	Input type for embedding
Options["dimensions"]	float64	Model-dependent	Override vector dimensions

Supported Models

Model	Default Dimensions
voyage-2	1024
voyage-large-2	1536
voyage-code-2	1536
voyage-lite-02-instruct	1024
voyage-3	1024
voyage-3-lite	512

Input Types

Voyage supports input type differentiation for asymmetric retrieval. Set the input_type option accordingly:

• document (default) — Use when embedding documents for storage
• query — Use when embedding queries for retrieval

// Embed documents for storage
docEmb, err := embedding.New("voyage", config.ProviderConfig{
    APIKey: os.Getenv("VOYAGE_API_KEY"),
    Options: map[string]any{
        "input_type": "document",
    },
})

// Embed queries for retrieval
queryEmb, err := embedding.New("voyage", config.ProviderConfig{
    APIKey: os.Getenv("VOYAGE_API_KEY"),
    Options: map[string]any{
        "input_type": "query",
    },
})

Direct Construction

import (
    voyageemb "github.com/lookatitude/beluga-ai/rag/embedding/providers/voyage"
)

emb, err := voyageemb.New(config.ProviderConfig{
    APIKey: os.Getenv("VOYAGE_API_KEY"),
    Model:  "voyage-code-2",
    Options: map[string]any{
        "input_type": "document",
    },
})
if err != nil {
    log.Fatal(err)
}

Code Embeddings

Voyage’s voyage-code-2 model is specifically tuned for code retrieval tasks:

emb, err := embedding.New("voyage", config.ProviderConfig{
    APIKey: os.Getenv("VOYAGE_API_KEY"),
    Model:  "voyage-code-2",
})
if err != nil {
    log.Fatal(err)
}

vectors, err := emb.Embed(ctx, []string{
    "func main() { fmt.Println(\"hello\") }",
    "def main(): print('hello')",
})