Bedrock Agent

This template demonstrates how to handle Amazon Bedrock Agent events using function-based actions.

Architecture

The template sets up:

1. Amazon Bedrock Agent: Orchestrates user requests and identifies tool usage.
2. AWS Lambda function: Resolves function calls using BedrockAgentFunctionResolver.
3. Amazon DynamoDB table: Stores and retrieves items used by the agent.

Code

• Function code: templates/agent
• Unit tests: tests/agent
• Infra stack: infra/stacks/agent.py

Deployment

Deploy the stack using:

mise run deploy agent

Features

• Function-based Actions: Uses the @app.tool() decorator to expose functions to Bedrock Agents.
• Automatic Parameter Mapping: Maps Bedrock Agent parameters directly to Python function arguments.
• Type Safety: Uses Pydantic models for data validation and camelCase alias conversion.
• Observability: Integrated with AWS Lambda Powertools for logging, tracing, and metrics.

Usage

Define Tools

Use the @app.tool() decorator to define tools that the agent can call:

@app.tool(name="getItem", description="Gets item details by ID")
def get_item(item_id: str) -> dict:
    return handler.get_item(item_id)

Item model

Field	Type	Description
id	UUID string	Unique item identifier (auto-generated, 1-50 chars)
name	string	Human-readable item name (1-100 chars)
description	string	Human-readable item description (max 500 chars)

Environment variables

Variable	Description	Required	Default
TABLE_NAME	DynamoDB table name	Yes	-
SERVICE_NAME	Powertools service name	No	bedrock-agent
METRICS_NAMESPACE	Powertools metrics namespace	No	BedrockAgent
LOG_LEVEL	Log level for the Lambda Logger	No	INFO