ChatBedrockConverse doesn't take into account boto3 setup_default_session. about langchain-aws HOT 3 OPEN

I believe you need to pass region_name directly into ChatBedrockConverse - or set the environment variable AWS_DEFAULT_REGION. boto3 and langchain_aws don't share variables, so you must specify the region name for the llm as well I believe. Let me know if that fixes your error.

from langchain-aws.

I believe you need to pass region_name directly into ChatBedrockConverse - or set the environment variable AWS_DEFAULT_REGION. boto3 and langchain_aws don't share variables, so you must specify the region name for the llm as well I believe. Let me know if that fixes your error.

Yes, setting region_name and credentials_profile_name in the ChatBedrockConverse constructor works.

Looking in the code of validate_environment in bedrock_converse.py, I see that a new boto3.Session object is being instantiated, and later, with that new session, the bedrock-runtime client is then created. Wouldn't it be better to just call boto3.client('bedrock-runtime') directly (or at least when no credentials_profile_name was provided) ? That way it will work with boto3.setup_default_session out of the box.

from langchain-aws.

Not a boto3 expert by any means - but if I understand correctly the code currently creates its own session, but you would like it to work with the session already setup? That sounds reasonable to me - if you want to open up a PR (with a corresponding test case) I would be happy to review. Lmk if I misunderstood

from langchain-aws.