Using boto3 session client instead of relying on env credentials file for APIs

[ventz]

talkwithhollis-langchain/app/worker.py

Lines 31 to 36 in 5066a90

	os.environ['AWS_ACCESS_KEY_ID'] = settings.aws_access_key_id
	os.environ['AWS_SECRET_ACCESS_KEY'] = settings.aws_secret_access_key
	self.llm = Bedrock(
	# Uncomment credentials_profile_name to use a profile defined in ~/.aws/credentials
	#credentials_profile_name=os.environ.get("AWS_BEDROCK_PROFILE_NAME", "talkwithhollis"),
	region_name=settings.aws_default_region,

Hi,

In some serverless environments (ie: some lambda/fargate/etc) this approach may fail due to the clobbering of path/environment, and tracking it down will be a pain.

You can use the boto3 session to pass an auth object like this:

import boto3

session = boto3.Session(aws_access_key_id=ACCESS_KEY, aws_secret_access_key=SECRET_KEY)
client = session.client("bedrock-runtime", region_name=settings.aws_default_region)

self.llm = BedrockChat(
            client=client,
            model_id=...
            ...
)

Also - note that if you go to BedrockChat (vs current Bedrock), you can utilize some of the new "chat only" models.

The new LangChain loads would be:

from langchain_community.chat_models import BedrockChat
from langchain_core.messages import HumanMessage, SystemMessage
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.output_parsers import StrOutputParser