AWS Agentic Boilerplate

A boilerplate for deploying multi-agent systems on AWS Lambda using the Strands Agents SDK.

The project implements an orchestrator pattern: a top-level agent receives user queries and delegates work to specialist agents, each equipped with their own tools. The whole stack runs as a single Lambda function behind API Gateway.

Architecture

flowchart TD
    GW[API Gateway] --> LH[Lambda Handler]
    LH --> O[Orchestrator Agent]
    O --> TA[Time Agent]
    O --> WA[Weather Agent]
    TA --> T[get_current_time]
    WA --> W[get_weather]

Loading

• Orchestrator — routes incoming queries to the right specialist(s) and combines their responses.
• Time Agent — answers date/time questions using the get_current_time tool.
• Weather Agent — answers weather questions using the get_weather tool (mock data).

Project Structure

.
├── lambda_handler.py          # Lambda entry point
├── agents/
│   ├── __init__.py            # Exports create_orchestrator
│   ├── config.py              # Model provider config (Bedrock / Anthropic)
│   ├── orchestrator.py        # Orchestrator agent
│   ├── specialists/
│   │   ├── __init__.py
│   │   ├── time_agent.py      # Time specialist agent
│   │   └── weather_agent.py   # Weather specialist agent
│   └── tools/
│       ├── __init__.py
│       ├── time.py            # get_current_time tool
│       └── weather.py         # get_weather tool
├── Dockerfile                 # Lambda container image
├── requirements.txt           # Python dependencies
└── .env.example               # Environment variable template

Prerequisites

• Python 3.13+
• An AWS account with Bedrock model access enabled (default), or an Anthropic API key
• Docker (for building the Lambda container image)

Configuration

Copy the env template and edit as needed:

cp .env.example .env

Model Provider

The project defaults to Amazon Bedrock — no API key required, authentication is handled by the Lambda execution role's IAM permissions.

To switch to Anthropic direct, uncomment Option B in agents/config.py and set ANTHROPIC_API_KEY in your environment.

Local Development

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

You can test the orchestrator directly in Python:

from agents import create_orchestrator

agent = create_orchestrator()
result = agent("What time is it?")
print(result)

Deployment

Build and push the container image, then create a Lambda function pointing at it:

docker build -t aws-agentic-boilerplate .

The Lambda handler is lambda_handler.handler.

API

The function expects an API Gateway proxy event with a JSON body:

{ "query": "What's the weather in Tokyo?" }

Response:

{ "answer": "Weather in Tokyo: sunny, 28°C, humidity 65%" }

Adding New Agents

1. Create a new tool in agents/tools/ using the @tool decorator.
2. Create a new specialist in agents/specialists/ that wraps an Agent with your tool.
3. Register the specialist in agents/orchestrator.py by adding it to the tools list.
4. Update the orchestrator's system prompt to describe when to use the new specialist.

See AGENTS.md for a deeper explanation of the agent architecture.

License

MIT