Voyage AI Provider

patelvivekdev/voyage-ai-provider is a community provider that uses Voyage AI to provide Embedding support for the AI SDK.

Setup

The Voyage provider is available in the voyage-ai-provider module. You can install it with

<Tabs items={['pnpm', 'npm', 'yarn', 'bun']}> <Tab> <Snippet text="pnpm add voyage-ai-provider" dark /> </Tab> <Tab> <Snippet text="npm install voyage-ai-provider" dark /> </Tab> <Tab> <Snippet text="yarn add voyage-ai-provider" dark /> </Tab> <Tab> <Snippet text="bun add voyage-ai-provider" dark /> </Tab> </Tabs>

Provider Instance

You can import the default provider instance voyage from voyage-ai-provider:

import { voyage } from 'voyage-ai-provider';

If you need a customized setup, you can import createVoyage from voyage-ai-provider and create a provider instance with your settings:

import { createVoyage } from 'voyage-ai-provider';

const voyage = createVoyage({
  // custom settings
});

You can use the following optional settings to customize the Voyage provider instance:

• baseURL string

  The base URL of the Voyage API. The default prefix is https://api.voyageai.com/v1.

• apiKey string

  API key that is being sent using the Authorization header. It defaults to the VOYAGE_API_KEY environment variable.

• headers Record<string,string>

  Custom headers to include in the requests.

• fetch (input: RequestInfo, init?: RequestInit) => Promise<Response>

  Custom fetch implementation. Defaults to the global fetch function. You can use it as a middleware to intercept requests, or to provide a custom fetch implementation for e.g. testing.

Text Embedding Models

You can create models that call the Voyage embeddings API using the .embeddingModel() factory method.

import { voyage } from 'voyage-ai-provider';

const embeddingModel = voyage.embeddingModel('voyage-3.5-lite');

You can use Voyage embedding models to generate embeddings with the embed or embedMany function:

import { voyage } from 'voyage-ai-provider';
import { embed } from 'ai';

const { embedding } = await embed({
  model: voyage.embeddingModel('voyage-3.5-lite'),
  value: 'sunny day at the beach',
  providerOptions: {
    voyage: {
      inputType: 'document',
    },
  },
});

Voyage embedding models support additional provider options that can be passed via providerOptions.voyage:

import { voyage } from 'voyage-ai-provider';
import { embed } from 'ai';

const { embedding } = await embed({
  model: voyage.embeddingModel('voyage-3.5-lite'),
  value: 'sunny day at the beach',
  providerOptions: {
    voyage: {
      inputType: 'query',
      outputDimension: 512,
    },
  },
});

The following provider options are available:

• inputType 'query' | 'document' | 'null'

  Specifies the type of input passed to the model. Defaults to 'null'.

  • 'null': When inputType is 'null', the embedding model directly converts the inputs into numerical vectors.

  For retrieval/search purposes it is recommended to use 'query' or 'document'.

  • 'query': The input is a search query, e.g., "Represent the query for retrieving supporting documents: ...".
  • 'document': The input is a document to be stored in a vector database, e.g., "Represent the document for retrieval: ...".

• outputDimension number

  The number of dimensions for the resulting output embeddings. Default is 'null'.

  • For example, voyage-code-3 and voyage-3-large support: 2048, 1024 (default), 512, and 256.
  • Refer to the model documentation for supported values.

• outputDtype 'float' | 'int8' | 'uint8' | 'binary' | 'ubinary'

  The data type for the output embeddings. Defaults to 'float'.

  • 'float': 32-bit floating-point numbers (supported by all models).
  • 'int8', 'uint8': 8-bit integer types (supported by voyage-3-large, voyage-3.5, voyage-3.5-lite, and voyage-code-3).
  • 'binary', 'ubinary': Bit-packed, quantized single-bit embedding values (voyage-3-large, voyage-3.5, voyage-3.5-lite, and voyage-code-3). The returned list length is 1/8 of outputDimension. 'binary' uses offset binary encoding.

  See FAQ: Output Data Types for more details.

• truncation boolean

  Whether to truncate the input texts to fit within the model's context length. If not specified, defaults to true.

You can find more models on the Voyage Library homepage.

Model Capabilities

Image Embedding

Example 1: Embed an image as a single embedding

import { voyage, ImageEmbeddingInput } from 'voyage-ai-provider';
import { embedMany } from 'ai';

const imageModel = voyage.imageEmbeddingModel('voyage-multimodal-3');

const { embeddings } = await embedMany<ImageEmbeddingInput>({
  model: imageModel,
  values: [
    {
      image:
        'https://raw.githubusercontent.com/voyage-ai/voyage-multimodal-3/refs/heads/main/images/banana_200_x_200.jpg',
    },
    {
      image: 'data:image/jpeg;base64,/9j/4AAQSkZJRgABAQAA...',
    },
  ],
  // or you can pass the array of images url and base64 string directly
  // values: [
  //   'https://raw.githubusercontent.com/voyage-ai/voyage-multimodal-3/refs/heads/main/images/banana_200_x_200.jpg',
  //   'data:image/jpeg;base64,/9j/4AAQSkZJRgABAQAA...',
  // ],
});

Example 2: Embed multiple images as single embedding

import { voyage, ImageEmbeddingInput } from 'voyage-ai-provider';
import { embedMany } from 'ai';

const imageModel = voyage.imageEmbeddingModel('voyage-multimodal-3');

const { embeddings } = await embedMany<ImageEmbeddingInput>({
  model: imageModel,
  values: [
    {
      image: [
        'https://raw.githubusercontent.com/voyage-ai/voyage-multimodal-3/refs/heads/main/images/banana_200_x_200.jpg',
        'data:image/jpeg;base64,/9j/4AAQSkZJRgABAQAA...',
      ],
    },
  ],
});

Multimodal Embedding

Example 1: Embed multiple texts and images as single embedding

import { voyage, MultimodalEmbeddingInput } from 'voyage-ai-provider';
import { embedMany } from 'ai';

const multimodalModel = voyage.multimodalEmbeddingModel('voyage-multimodal-3');

const { embeddings } = await embedMany<MultimodalEmbeddingInput>({
  model: multimodalModel,
  values: [
    {
      text: ['Hello, world!', 'This is a banana'],
      image: [
        'https://raw.githubusercontent.com/voyage-ai/voyage-multimodal-3/refs/heads/main/images/banana_200_x_200.jpg',
      ],
    },
    {
      text: ['Hello, coders!', 'This is a coding test'],
      image: ['data:image/jpeg;base64,/9j/4AAQSkZJRgABAQAA...'],
    },
  ],
});

The following constraints apply to the values list:

• The list must not contain more than 1,000 values.
• Each image must not contain more than 16 million pixels or be larger than 20 MB in size.
• With every 560 pixels of an image being counted as a token, each input in the list must not exceed 32,000 tokens, and the total number of tokens across all inputs must not exceed 320,000.

Voyage multimodal embedding models support additional provider options that can be passed via providerOptions.voyage:

import { voyage, MultimodalEmbeddingInput } from 'voyage-ai-provider';
import { embedMany } from 'ai';

const multimodalModel = voyage.multimodalEmbeddingModel('voyage-multimodal-3');

const { embeddings } = await embedMany<MultimodalEmbeddingInput>({
  model: multimodalModel,
  values: [
    {
      text: ['Hello, world!'],
      image: ['data:image/jpeg;base64,/9j/4AAQSkZJRgABAQAA...'],
    },
  ],
  providerOptions: {
    voyage: {
      inputType: 'query',
      outputEncoding: 'base64',
      truncation: true,
    },
  },
});

The following provider options are available:

• inputType 'query' | 'document'

  Specifies the type of input passed to the model. Defaults to 'query'.

  When inputType is specified as 'query' or 'document', Voyage automatically prepends a prompt to your inputs before vectorizing them, creating vectors tailored for retrieval/search tasks:

  • 'query': Prepends "Represent the query for retrieving supporting documents: "
  • 'document': Prepends "Represent the document for retrieval: "

• outputEncoding 'base64'

  The data encoding for the resulting output embeddings. Defaults to null (list of 32-bit floats).

  • If null, embeddings are returned as a list of floating-point numbers (float32).
  • If 'base64', embeddings are returned as a Base64-encoded NumPy array of single-precision floats.

  See FAQ: Output Data Types for more details.

• truncation boolean

  Whether to truncate the inputs to fit within the model's context length. If not specified, defaults to true.

Model Capabilities