In this post we will review the step required to create an AWS bderock agent that run a lambda function. I am creating this post since AWS documentation is not sufficient to achieve this goal. I guess most software engineers today assume that you will use repeatadly use LLM such as ChatGPT to solve their no-so-good product and documentation issues.
Create an Agent
Start with agent creation, you can follow the steps in this post.
Create Action Group
Notice:
Make sure to select a model that you have permissions for. For example, in case you choose one of Anthropic models, you will probably fail with permissions since you did not purchased this model. And yes, this error will be just a general error forcing you to look for the actual failure reason yourself.
Next we need to create an action group:
- Edit the agent, and under the action group section click add new action group
- Select "Define with API schemas" as the action group type
- Select "Quick create a new Lambda function" in the action group invocation
- Select "Define via in-line schema editor" in the action group schema
An example schema is the following:
openapi: 3.0.0
info:
title: Person Info API
version: 1.0.0
description: API to get the person info
paths:
/get_person_info:
get:
summary: Gets the person info
description: Gets the person info
operationId: get_person_info
responses:
'200':
description: Gets the person info
content:
'application/json':
schema:
type: object
properties:
info:
type: string
description: The person info
Create the Lambda
After clicking save, we can click on the button "View" next to the "Select Lambda function" combo. This opens a new tab where we can edit the lambda code. An example working lambda code is:
import logging
from typing import Dict, Any
from http import HTTPStatus
import json
logger = logging.getLogger()
logger.setLevel(logging.INFO)
def lambda_handler(event: Dict[str, Any], context: Any) -> Dict[str, Any]:
try:
action_group = event['actionGroup']
api_path = event['apiPath']
http_method = event['httpMethod']
message_version = event.get('messageVersion', '1.0')
# Your business logic output
result = {
"info": "this Person ate the entire cake"
}
response = {
"messageVersion": message_version,
"response": {
"actionGroup": action_group,
"apiPath": api_path,
"httpMethod": http_method,
"httpStatusCode": 200,
"responseBody": {
"application/json": {
"body": json.dumps(result)
}
}
}
}
logger.info("Response: %s", response)
return response
except Exception as e:
logger.error("Error: %s", str(e))
return {
"messageVersion": "1.0",
"response": {
"actionGroup": event.get("actionGroup", ""),
"apiPath": event.get("apiPath", ""),
"httpMethod": event.get("httpMethod", ""),
"httpStatusCode": 500,
"responseBody": {
"application/json": {
"body": json.dumps({"error": str(e)})
}
}
}
}
Notice:
Do not use the lambda code example that you get, it contains bugs.
Using the Lambda
Using the lambda is not straight forward. You need to perform additional 3 steps:
1. Deploy
After each update the the lambda code, click on Deploy.
Then, on the top of the screen, click on Actions, Publish new version.
2. Permissions
You need to add permissions to the agent role. Notice the permission includes the deployed version, so you need to rerurn this after each deployment. An example for permissions update is:
aws lambda add-permission \
--function-name get_person_info-3urcv \
--statement-id allow-bedrock-invoke-version-1 \
--action lambda:InvokeFunction \
--principal bedrock.amazonaws.com \
--source-arn arn:aws:bedrock:us-east-1:123460611234:agent/* \
--qualifier 2 \
--region us-east-1
3. Configure the agent
Edit the agent, then edit the action group, and replace the used lambda version under "Select Lambda function". Then save AND prepare the agent to use the updated configuration.
Final Note
Fun, it was not, but eventually it is working. We can now chat with the agent from the AWS console and using code like the one in the agent creation post.
