FastAgency: Getting Started with Multi-Agent Workflows in Minutes

Getting Started with FastAgency

FastAgency makes it incredibly easy to deploy multi-agent workflows built with AG2 (formerly AutoGen). In this tutorial, you will set up a FastAgency project and launch a web-based multi-agent chat application in under 10 minutes.

Prerequisites

Python 3.10 or later
An OpenAI API key (or any LLM API key supported by AG2)
pip installed

Step 1: Install Cookiecutter and Scaffold Your Project

pip install cookiecutter

cookiecutter https://github.com/ag2ai/cookiecutter-fastagency.git

Follow the interactive prompts. Choose fastapi+mesop as your app type for a web-based chat UI with a REST API backend. The scaffold creates a complete project structure with a devcontainer, default workflow, and all dependencies.

Step 2: Set Your API Key

export OPENAI_API_KEY=sk-your-key-here

Step 3: Understand the Default Workflow

The generated workflow.py defines a simple two-agent conversation using AG2's ConversableAgent. The workflow is wrapped in FastAgency's Workflow class, which provides the unified interface for deployment across different UIs.

Step 4: Run Your Application

cd my_fastagency_app

fastagency run

This starts a local web server. Open your browser to the displayed URL (typically http://localhost:8000) and start chatting with your multi-agent system.

Step 5: Run the Tests

pytest -s

The scaffold includes a test suite that validates your workflow. FastAgency's Tester class makes it easy to add more tests as your agent logic grows.

Deploying to Production

When you are ready to scale, FastAgency's network adapters let you switch from local development to production with minimal code changes. Use the FastAPIAdapter to serve your workflow as a REST API, or combine it with the NatsAdapter for distributed, multi-worker deployments.

Conclusion

FastAgency removes the friction between prototyping and production for multi-agent AI workflows. With cookiecutter scaffolding, a unified API, and scalable deployment options, it is the fastest way to bring your AG2 agents to the real world.