AutoGen
Microsoft's multi-agent framework — now in maintenance mode, succeeded by Microsoft Agent Framework.
About
Microsoft's open-source framework for building multi-agent AI applications, where specialized agents collaborate through conversations, tool use, and code execution. Layered Core/AgentChat/Extensions architecture for Python and .NET, plus a no-code AutoGen Studio GUI.
As of 2026 it is officially in maintenance mode — bug/security fixes only, community-managed — with Microsoft Agent Framework (1.0, enterprise-ready) as its successor. Existing codebases are fine; new projects should start with the successor (migration guide on learn.microsoft.com).
What it can do
- Multi-agent conversation patterns: two-agent chat, group chat, round-robin/selector orchestration
- Code-executing agents (local, Docker, Jupyter, Azure)
- MCP tool integration via McpWorkbench; cross-language runtime (Python + .NET)
- Local and distributed runtimes via gRPC
- AutoGen Studio no-code GUI for prototyping agent teams
- AutoGen Bench benchmarking suite for evaluating agent performance
Setup
- Install Python 3.10+ and run
pip install -U "autogen-agentchat" "autogen-ext[openai]". - Export your model provider key (e.g.
OPENAI_API_KEY). - Create agents (e.g.
AssistantAgentwithOpenAIChatCompletionClient) and define their roles. - Compose teams: two-agent chats, group chats, or selector/round-robin workflows; attach tools or MCP workbenches (MCP verified in README).
- (Optional)
pip install autogenstudioand runautogenstudio ui --port 8080for the no-code Studio.
Prerequisites
Python 3.10+; a model-provider API key (e.g. OpenAI/Azure); Docker optional for sandboxed code execution.
Use cases
Works with
Trust score — 7/10
Official Microsoft project with massive adoption (61k+ stars) and excellent docs, so the code itself is trustworthy. Docked hard for longevity: the README explicitly states maintenance mode (bug/security fixes only, community-managed) and directs new users to Microsoft Agent Framework. Recommending it for new builds without the successor caveat would be misleading.
★ 61,374 GitHub stars
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.7/10
- Maintenance · recency, releases, issue load
- 4.2/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 6.0/10
- · 61,374 GitHub stars (log-scaled vs catalog peers)
- · blended with ~379,923 monthly package downloads
- · last push 180d ago; 0 releases/12mo; 1103 open issues
- · manual review: maintenance mode; Microsoft Agent Framework is the successor
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.