AutoGen

The microsoft/autogen README states plainly: "AutoGen is now in maintenance mode. It will not receive new features or enhancements and is community managed going forward," and directs new users to Microsoft Agent Framework (MAF), which shipped as a production-ready 1.0 for Python and .NET on April 3, 2026. This isn't a rumor or a stale changelog — it's the current text at the top of the repo's own README. Practically, this also explains a gap in the release history: the last PyPI release of autogen-agentchat (v0.7.5) was uploaded 2025-09-30, and while commits to the main branch continued through 2026-04-06, there has been no new tagged release since — consistent with the README's statement that contributions are now limited to bug fixes, security patches, and docs.

In late 2024, AutoGen's original creators (including Chi Wang and Qingyun Wu) left Microsoft and forked the project into AG2 (github.com/ag2ai/ag2), taking over the autogen and pyautogen PyPI package names and the original Discord community. AG2 kept the older, conversational-agent API style (AutoGen v0.2) and is independently and actively developed — 4,881 GitHub stars, last pushed 2026-08-20 at time of writing. Microsoft, meanwhile, rewrote microsoft/autogen into an event-driven, actor-based architecture (v0.4, released early 2025) under new package names (autogen-agentchat, autogen-core), and has since folded that lineage together with Semantic Kernel into Microsoft Agent Framework, which had 13,008 stars and was last pushed 2026-08-20. All three answer to some form of "AutoGen" in search results and comparison articles; they are not interchangeable, and only two of the three (AG2, Agent Framework) are receiving active feature development.

Setting the lifecycle question aside, the v0.4 architecture microsoft/autogen shipped is a genuinely well-regarded design: an actor-model, event-driven core (autogen-core) with a higher-level conversational API on top (autogen-agentchat), separating low-level message-passing agents from the multi-agent "team" patterns (round-robin chat, selector chat, swarm) most builders actually want. AutoGen Studio ships in the same repo as a no-code GUI for prototyping these workflows (pip install autogenstudio), explicitly scoped by its own docs as a rapid-prototyping tool rather than a production deployment surface. The repo carries 60,545 GitHub stars and 9,129 forks, reflecting the reach the original project had before the split.

Code is MIT-licensed; non-code content (docs, etc.) is under CC BY 4.0 — both confirmed in the repository's LICENSE and LICENSE-CODE files. No open security advisories were found for the autogen-agentchat, autogen-core, or pyautogen PyPI package names in either the OSV.dev database or GitHub's Security Advisories database as of this check. Vulnerability reporting goes through Microsoft's standard MSRC process (a generic template shared across all Microsoft repos, not a dedicated AutoGen trust/compliance page).

AutoGen is free and open source (MIT-licensed code); there is no vendor-hosted paid tier or subscription. Running it costs whatever compute and LLM API usage (e.g., OpenAI, Azure OpenAI) you point it at.