AG2

AG2 is the open-source multi-agent framework that grew out of Microsoft AutoGen. On November 11, 2024, AutoGen's original creators, Chi Wang and Qingyun Wu, announced they were evolving the project into AG2 under a new, independently governed AG2ai organization, forking from microsoft/autogen at v0.2.35. The project switched to the Apache 2.0 license from v0.3 and ships on PyPI as ag2 (with autogen and pyautogen aliases).

The framework's building blocks are conversable agents: assistant agents that call LLMs, user-proxy agents that can execute code or ask a human for input, and group-chat managers that route multi-agent conversations. It also bundles patterns for sequential chats, nested chats, and tool use with code execution in sandboxes. Microsoft's original AutoGen repository has since gone into maintenance mode while Microsoft directs new agent development toward its Agent Framework, so AG2 is now the living continuation of the AutoGen lineage.

1. Install Python 3.9+ and run pip install ag2.
2. Configure an LLM endpoint in an LLMConfig (OpenAI, Azure, or local models).
3. Create assistant and user-proxy agents, register tools/functions.
4. Run a conversation with initiate_chat or a group chat manager.