AI Agents Hub
← Back to all agents
Framework Open Source
AutoGen
Microsoft’s framework for building conversational multi-agent systems.
Our take
Research-flavored and powerful, but steeper learning curve than CrewAI.
What it is
AutoGen is Microsoft's framework for building multi-agent systems where agents converse with each other to solve a task. Each agent can have its own system prompt, tools, and human-in-the-loop behavior. It is research-flavored and used heavily in academic papers on multi-agent collaboration.
Best for
- Research projects on multi-agent collaboration
- Custom agent systems with human-in-the-loop
- Conversational workflows with branching dialogues
Pros
- Powerful and flexible — widely used in research papers
- Microsoft-backed, mature codebase
- Strong human-in-the-loop patterns
Cons
- Steeper learning curve than CrewAI
- Documentation can feel scattered across versions
Quick start
- Install: `pip install pyautogen` (or `autogen-agentchat`).
- Set your `config_list = [{"model": "gpt-4o", "api_key": "..."}]`.
- Create agents: `AssistantAgent` for the LLM and `UserProxyAgent` to execute code / human input.
- Spin up a GroupChat if you want 3+ agents collaborating; otherwise `.initiate_chat(assistant, message=...)` for one-to-one.
- Enable Docker in `code_execution_config` if your agents will run code, otherwise use a sandboxed work_dir.
Sample input / output
Input
Task: "Write a Python script that downloads the top 10 trending repos from GitHub and saves them to CSV. Test it."Output
Assistant generates the script (60+ lines, uses requests + pandas), explains the logic, then runs it via UserProxyAgent:
Verified: 10 repos saved to top_repos.csv. Passes a quick AST lint. AutoGen's enable_docker was left off so the script ran inside a sandboxed work_dir.Benchmarks
GitHub stars 38,000+ — github.com/microsoft/autogen
License MIT + Creative Commons for code-execution docs
Successor AutoGen functionality merged into Microsoft Agent Framework (Oct 2026) — Microsoft Devblogs
Patterns AssistantAgent / UserProxyAgent / GroupChat / Nested chat / Docker-isolated execution
Pricing
Free library
Underlying models
Self-hosting
Yes — AutoGen can be self-hosted under the MIT license. This gives you full control over data and deployment.
Should you pick this?
Pick it if You want Microsoft's research-grade multi-agent patterns with deep human-in-the-loop and code-execution safety — especially for academic or open-ended conversational workflows.
Skip it if You want a quicker on-ramp — CrewAI is the most approachable. For production graph workflows with cycles, LangGraph is more mature. AutoGen is now part of Microsoft Agent Framework going forward.