langchain-ai/langchain-aws

Build LangChain Applications on AWS

What it solves

This project provides a dedicated set of integrations for Amazon Web Services (AWS) within the LangChain and LangGraph ecosystems. It simplifies the process of building AI agents and LLM applications by providing a unified interface to access AWS-specific AI services like Bedrock, SageMaker, and various AWS database and memory services.

How it works

The repository is a monorepo containing several packages that bridge the gap between AWS services and LangChain/LangGraph:

  • LLMs and Agents: Integrates AWS Bedrock and SageMaker Endpoints for model inference, and supports Bedrock Agents for agentic workflows.
  • Vector Stores and Retrievers: Connects to Amazon MemoryDB, S3 Vectors, ElastiCache for Valkey, and Amazon Kendra for RAG (Retrieval-Augmented Generation) capabilities.
  • State Management: Provides custom checkpointers and memory stores for LangGraph using DynamoDB, ElastiCache Valkey, and Bedrock AgentCore Memory to maintain agent state and long-term memory.
  • Code Execution: Includes a sandbox backend for secure code execution via Amazon Bedrock AgentCore Code Interpreter in isolated MicroVM environments.

Who it’s for

Developers building LLM-powered applications and AI agents who are already using or plan to use the AWS cloud infrastructure for their models, data storage, and agent state management.

Highlights

  • Comprehensive AWS Integration: Covers everything from model inference (Bedrock/SageMaker) to vector databases and retrieval systems.
  • LangGraph State Persistence: Dedicated AWS-native checkpointers for DynamoDB and ElastiCache to handle agent memory.
  • Secure Code Execution: Integration with Bedrock AgentCore for isolated code interpreter sandboxes.
  • **Bedrock AgentCore Tools: Built-in toolkits for browser automation and code execution that can be used directly with LangChain agents.