langchain-voyageai

This package contains the LangChain integrations for VoyageAI through their voyageai client package.

Installation and Setup

• Install the LangChain partner package

pip install langchain-voyageai

• Get an VoyageAI api key and set it as an environment variable (VOYAGE_API_KEY) or use the API key as a parameter in the Client.

Text Embedding Model

See a usage example

from langchain_voyageai import VoyageAIEmbeddings

Voyage-4 Family Models

The latest generation of VoyageAI embedding models with improved quality and flexibility:

# voyage-4: Balanced model for general-purpose and multilingual retrieval
embeddings = VoyageAIEmbeddings(model="voyage-4")

# voyage-4-lite: Optimized for latency and cost, highest batch throughput (1M tokens/batch)
embeddings = VoyageAIEmbeddings(model="voyage-4-lite")

# voyage-4-large: Best retrieval quality for demanding applications
embeddings = VoyageAIEmbeddings(model="voyage-4-large")

All voyage-4 family models support flexible output dimensions (256, 512, 1024, 2048):

embeddings = VoyageAIEmbeddings(
    model="voyage-4",
    output_dimension=512  # Choose from: 256, 512, 1024, 2048
)

Code Embeddings

voyage-code-4 is optimized for code retrieval and supports flexible output dimensions (256, 512, 1024, 2048):

# voyage-code-4: optimized for code retrieval
embeddings = VoyageAIEmbeddings(model="voyage-code-4")

embeddings = VoyageAIEmbeddings(
    model="voyage-code-4",
    output_dimension=512  # Choose from: 256, 512, 1024, 2048
)

Contextualized Chunk Embeddings

voyage-context-4 produces contextualized chunk embeddings, where each chunk is embedded in the context of the other chunks from the same document:

# voyage-context-4: contextualized chunk embeddings for document-aware retrieval
embeddings = VoyageAIEmbeddings(model="voyage-context-4")

It also supports flexible output dimensions (256, 512, 1024, 2048):

embeddings = VoyageAIEmbeddings(
    model="voyage-context-4",
    output_dimension=512  # Choose from: 256, 512, 1024, 2048
)

Custom API Endpoint

You can specify a custom API endpoint using the base_url parameter:

embeddings = VoyageAIEmbeddings(
    model="voyage-3.5",
    base_url="https://ai.mongodb.com/v1"
)

This is useful for MongoDB Atlas users or custom deployments.