AutoGen

🎯 The Problem It Solves

A single LLM call can answer a question; a *team* of agents with different skills can solve problems that require reasoning, verification, and iteration. AutoGen (Microsoft) gives you exactly that: a framework where you spawn agents with different roles (User, Assistant, Critic, CodeExecutor), define conversation patterns, and let them propose, critique, and execute code until a solution converges. It's the go-to for academic research on multi-agent collaboration and increasingly common in production too.

🔧 How It Works

AutoGen is a Python (and now .NET) library from Microsoft Research. You define agents (each wrapping an LLM and optionally a tool like a Python REPL), then orchestrate conversations between them (two-agent chat, group chat with a speaker-selection agent, nested chat). Agents can write and execute code in a sandboxed container (Docker or Jupyter), making AutoGen unusually strong for tasks that need computation — math, data analysis, algorithm design. v0.4 introduced a fully async, event-driven rewrite with better extensibility. It integrates with any OpenAI-compatible endpoint.

🚀 Installation & Quick Start

pip install autogen-agentchat
pip install autogen-ext  # for extensions (OpenAI, Azure, etc.)
# For code execution:
pip install autogen-ext[docker]