AutoGen is a Python framework, originating from Microsoft Research, for building multi-agent AI applications in which LLM-backed agents act autonomously or alongside humans. Agents exchange messages, call tools (including MCP servers), and can be composed into orchestrations where one assistant delegates to specialized expert agents. It requires Python 3.10 or later and, in the quickstart examples, drives models such as OpenAI's via an extension client.
The framework is deliberately layered so it can be used at different levels of abstraction:
- Core API — message passing, event-driven agents, and local/distributed runtime, with cross-language support for Python and .NET
- AgentChat API — a simpler, opinionated high-level API for rapid prototyping and common patterns like two-agent and group chats
- Extensions API — first- and third-party add-ons providing model clients (e.g. OpenAI, Azure OpenAI) and capabilities such as code execution
Two developer tools round out the ecosystem: AutoGen Studio, a no-code GUI for prototyping multi-agent workflows (explicitly not production-ready), and AutoGen Bench, a benchmarking suite for evaluating agent performance.
The project is now officially in maintenance mode: it is community-managed, receives only bug fixes, security patches, and documentation updates, and no new features. Microsoft directs new projects to its successor, the Microsoft Agent Framework, and provides a migration guide for existing AutoGen users. Anyone evaluating AutoGen should weigh that maintenance status against the framework's design when choosing it for new work.
