How to Intermediate · 4 min read

How to use Bedrock embeddings with LangChain

Quick answer

Use the AWS Bedrock embeddings by calling the Bedrock runtime via boto3 to generate embeddings, then wrap this call in a custom LangChain embedding class implementing embed_documents and embed_query. This enables LangChain to use Bedrock embeddings for vector search and retrieval workflows.

PREREQUISITES

Python 3.8+
AWS credentials configured with access to Bedrock
pip install boto3 langchain

Setup

Install the required packages and configure AWS credentials to access Bedrock. You need boto3 for AWS API calls and langchain for vector store integration.

Install packages: pip install boto3 langchain
Configure AWS credentials via ~/.aws/credentials or environment variables AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY.

bash

pip install boto3 langchain

Step by step

Create a custom LangChain embedding class that calls AWS Bedrock's invoke_model API to generate embeddings. Use the amazon.titan-text-embedding model or similar. Then use this embedding class with LangChain vector stores.

python

import os
import boto3
from langchain.embeddings.base import Embeddings

class BedrockEmbeddings(Embeddings):
    def __init__(self, region_name="us-east-1"):
        self.client = boto3.client("bedrock-runtime", region_name=region_name)
        self.model_id = "amazon.titan-text-embedding"  # Confirm current embedding model ID

    def embed_documents(self, texts):
        embeddings = []
        for text in texts:
            response = self.client.invoke_model(
                modelId=self.model_id,
                body=text.encode("utf-8")
            )
            # The response body is a streaming HTTP response, read and decode
            embedding_bytes = response['Body'].read()
            # The embedding is returned as JSON bytes, decode and parse
            import json
            embedding_json = json.loads(embedding_bytes)
            embeddings.append(embedding_json["embedding"])
        return embeddings

    def embed_query(self, text):
        return self.embed_documents([text])[0]


# Usage example
if __name__ == "__main__":
    embeddings = BedrockEmbeddings(region_name="us-east-1")
    texts = ["Hello world", "LangChain with Bedrock embeddings"]
    vectors = embeddings.embed_documents(texts)
    print("Embeddings:", vectors)

output

Embeddings: [[0.01, 0.02, ...], [0.03, 0.04, ...]]  # Example float vectors

Common variations

You can customize the region_name for Bedrock, use async calls with aioboto3, or integrate with different LangChain vector stores like FAISS or Chroma. Also, update modelId if AWS releases new embedding models.

python

import aioboto3
import asyncio
from langchain.embeddings.base import Embeddings

class AsyncBedrockEmbeddings(Embeddings):
    def __init__(self, region_name="us-east-1"):
        self.region_name = region_name
        self.model_id = "amazon.titan-text-embedding"

    async def embed_documents(self, texts):
        embeddings = []
        async with aioboto3.client("bedrock-runtime", region_name=self.region_name) as client:
            for text in texts:
                response = await client.invoke_model(
                    modelId=self.model_id,
                    body=text.encode("utf-8")
                )
                embedding_bytes = await response['Body'].read()
                import json
                embedding_json = json.loads(embedding_bytes)
                embeddings.append(embedding_json["embedding"])
        return embeddings

    async def embed_query(self, text):
        return (await self.embed_documents([text]))[0]

# Async usage example
async def main():
    embeddings = AsyncBedrockEmbeddings()
    vectors = await embeddings.embed_documents(["Async example"])
    print("Async embeddings:", vectors)

if __name__ == "__main__":
    asyncio.run(main())

output

Async embeddings: [[0.01, 0.02, ...]]  # Example float vectors

Troubleshooting

If you get AccessDeniedException, verify your AWS credentials and Bedrock permissions.
If the response body is empty or malformed, check the modelId and ensure the Bedrock endpoint is correct.
For timeout errors, increase client timeout or check network connectivity.

✅

Key Takeaways

Use boto3 to call AWS Bedrock invoke_model for embeddings.
Wrap Bedrock calls in a LangChain Embeddings subclass for seamless integration.
Confirm the current Bedrock embedding model ID, e.g., amazon.titan-text-embedding.
Async embedding calls are possible with aioboto3 for better performance.
Check AWS permissions and network settings if you encounter errors.

Verified 2026-04 · amazon.titan-text-embedding

Verify ↗