LangChain AWS Integrations

Warnings

• breaking The `langchain-aws` library was split from the main `langchain` package in late 2023 / early 2024. All AWS-specific integrations (e.g., Bedrock, SageMaker) now reside in `langchain-aws` and require explicit installation and updated import paths.
  Fix: Migrate import paths from `from langchain.<module> import <Class>` to `from langchain_aws.<module> import <Class>`. For example, `from langchain.llms import Bedrock` becomes `from langchain_aws.llms import BedrockLLM`.
• gotcha Proper configuration of AWS credentials and region is essential. `langchain-aws` relies on `boto3`'s default credential chain. Misconfiguration can lead to `ClientError` (e.g., 'NotAuthorizedException', 'UnrecognizedClientException').
  Fix: Ensure `AWS_REGION_NAME` is set (e.g., as an environment variable or via `~/.aws/config`). Provide credentials via environment variables (`AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`), shared credential files (`~/.aws/credentials`), or IAM roles/profiles. Refer to `boto3` documentation for details.
• gotcha Bedrock model IDs are specific and may vary by AWS region or API availability. Using an incorrect or unavailable model ID will result in runtime errors from Bedrock.
  Fix: Verify the exact model ID (e.g., `anthropic.claude-3-sonnet-20240229-v1:0`) is available in your specified `region_name`. You can check available models through the AWS Bedrock console or programmatically via `boto3`.
• gotcha LangChain differentiates between chat models (`ChatBedrock`) and text completion models (`BedrockLLM`). Their interfaces and expected input/output formats (e.g., `HumanMessage` for chat, string for LLM) differ. Using the wrong class or input format can cause unexpected behavior or errors.
  Fix: Use `ChatBedrock` for models designed for conversational turns (e.g., Claude, Llama 2 Chat) and provide inputs as `list[BaseMessage]`. Use `BedrockLLM` for models primarily designed for text generation from a single prompt string.

Install

• pip install langchain-aws Install `langchain-aws`

Imports

• ChatBedrock

  from langchain_aws.chat_models import ChatBedrock

  As of LangChain 0.1.0 (and subsequent versions), AWS integrations moved to `langchain-aws`.
• BedrockLLM

  from langchain_aws.llms import BedrockLLM

  Old Bedrock LLM class was directly in `langchain` before the package split.
• BedrockEmbeddings

  from langchain_aws.embeddings import BedrockEmbeddings

  Embeddings integrations for AWS are now part of `langchain-aws`.

Quickstart

This quickstart demonstrates how to initialize `ChatBedrock` and use it to invoke a large language model. It requires the `AWS_REGION_NAME` environment variable to be set, and valid AWS credentials configured for `boto3` to access Amazon Bedrock.

import os
from langchain_aws.chat_models import ChatBedrock
from langchain_core.messages import HumanMessage

# Ensure AWS_REGION_NAME environment variable is set
# and AWS credentials are configured (e.g., via ~/.aws/credentials or env vars)

region = os.environ.get("AWS_REGION_NAME", "us-east-1")
model_id = "anthropic.claude-3-sonnet-20240229-v1:0" # Example Bedrock model ID

try:
    llm = ChatBedrock(
        model_id=model_id,
        region_name=region
    )

    messages = [
        HumanMessage(
            content="Tell me a short story about a brave knight and a wise dragon."
        )
    ]

    print(f"Invoking {model_id} in {region}...")
    response = llm.invoke(messages)
    print("\n--- Response ---")
    print(response.content)

except Exception as e:
    print(f"An error occurred: {e}")
    print("Please ensure AWS_REGION_NAME is set and AWS credentials are configured.")
    print("Example: export AWS_REGION_NAME=us-east-1")
    print("See https://boto3.amazonaws.com/v1/documentation/api/latest/guide/credentials.html")