A programming framework for agentic AI
AutoGen is a framework for creating multi-agent AI applications that can act autonomously or work alongside humans.
[!CAUTION] ⚠️ Maintenance Mode
AutoGen is now in maintenance mode. It will not receive new features or enhancements and is community managed going forward.
New users should start with Microsoft Agent Framework. Existing users are encouraged to migrate using the AutoGen → Microsoft Agent Framework migration guide.
Microsoft Agent Framework (MAF) is the enterprise‑ready successor to AutoGen. Microsoft Agent FrameworkAF in now available as a production-ready release: stable APIs, and a commitment to long-term support. Whether you're building a single assistant or orchestrating a fleet of specialized agents, Microsoft Agent Framework 1.0 gives you enterprise-grade multi-agent orchestration, multi-provider model support, and cross-runtime interoperability via A2A and MCP.
AutoGen requires Python 3.10 or later.
# Install AgentChat and OpenAI client from Extensions
pip install -U "autogen-agentchat" "autogen-ext[openai]"The current stable version can be found in the releases. If you are upgrading from AutoGen v0.2, please refer to the Migration Guide for detailed instructions on how to update your code and configurations.
# Install AutoGen Studio for no-code GUI
pip install -U "autogenstudio"The following samples call OpenAI API, so you first need to create an account and export your key as export OPENAI_API_KEY="sk-...".
Create an assistant agent using OpenAI's GPT-4o model. See other supported models.
import asyncio
from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient
async def main() -> None:
model_client = OpenAIChatCompletionClient(model="gpt-4.1")
agent = AssistantAgent("assistant", model_client=model_client)
print(await agent.run(task="Say 'Hello World!'"))
await model_client.close()
asyncio.run(main())Create a web browsing assistant agent that uses the Playwright MCP server.
…Warning: Only connect to trusted MCP servers as they may execute commands in your local environment or expose sensitive information.
You can use AgentTool to create a basic multi-agent orchestration setup.
…For more advanced multi-agent orchestrations and workflows, read AgentChat documentation.
Use AutoGen Studio to prototype and run multi-agent workflows without writing code.
Caution: AutoGen Studio is meant to help you rapidly prototype multi-agent workflows and demonstrate an example of end user interfaces built with AutoGen. It is not meant to be a production-ready app. Developers are encouraged to use the AutoGen framework to build their own applications, implementing authentication, security and other features required for deployed applications. See the security note for more details.
# Run AutoGen Studio on http://localhost:8080
autogenstudio ui --port 8080 --appdir ./my-appPioneered in Microsoft Research, AutoGen opened the door to experimental multi-agent orchestration patterns that inspired the community. While AutoGen is now in maintenance mode, existing users can continue to use the framework with the architecture described below. For new projects, we recommend Microsoft Agent Framework, which builds on the lessons learned from AutoGen with enterprise-grade support.
The autogen framework uses a layered and extensible design. Layers have clearly divided responsibilities and build on top of layers below. This design enables you to use the framework at different levels of abstraction, from high-level APIs to low-level components.
The ecosystem also supports two essential developer tools:
You can use the AutoGen framework and developer tools to create applications for your domain. For example, Magentic-One is a state-of-the-art multi-agent team built using AgentChat API and Extensions API that can handle a variety of tasks that require web browsing, code execution, and file handling.
For community support, visit our Discord server or GitHub Discussions. Note that AutoGen is now community-managed and responses may be limited.
Starting a new project? Head to Microsoft Agent Framework for the latest multi-agent capabilities with long-term support.
Existing AutoGen user? Use the migration guide to transition, or refer to the resources below for current AutoGen documentation.
Interested in contributing? See CONTRIBUTING.md for guidelines. As AutoGen is in maintenance mode, contributions are limited to bug fixes, security patches, and documentation improvements. For feature development, consider contributing to Microsoft Agent Framework.
Have questions? Check out our Frequently Asked Questions (FAQ) for answers to common queries. Community support is available through GitHub Discussions and the Discord server, though response times may vary as AutoGen is now community-managed. For actively supported tooling, see Microsoft Agent Framework.
Microsoft and any contributors grant you a license to the Microsoft documentation and other content in this repository under the Creative Commons Attribution 4.0 International Public License, see the LICENSE file, and grant you a license to any code in the repository under the MIT License, see the LICENSE-CODE file.
Microsoft, Windows, Microsoft Azure, and/or other Microsoft products and services referenced in the documentation may be either trademarks or registered trademarks of Microsoft in the United States and/or other countries. The licenses for this project do not grant you rights to use any Microsoft names, logos, or trademarks. Microsoft's general trademark guidelines can be found at http://go.microsoft.com/fwlink/?LinkID=254653.
Privacy information can be found at https://go.microsoft.com/fwlink/?LinkId=521839
Microsoft and any contributors reserve all other rights, whether under their respective copyrights, patents, or trademarks, whether by implication, estoppel, or otherwise.
[Bug]: MemoryContent.mime_type enum degenerates to str after dump/load roundtrip (ListMemory)
Feature: Cryptographic action receipts for enterprise agent governance (AAR)
gaia_question_scorer vacuously scores a degenerate/empty agent answer as correct when the ground truth normalizes to empty
FunctionTool(strict=True) loses strict across dump_component / load_component
Multi-agent systems need a 'mission keeper' role — not a Boss Agent, but a dedicated goal integrity node
Handoff auto-name with spaces builds OK but fails in convert_tools / AssistantAgent.run
TextMentionTermination(sources=...) drops sources on dump_component / load_component
OpenAIChatCompletionClient.create_stream returns finish_reason "unknown" when the finish chunk also carries usage (seen with GPT-6 Sol/Luna/Astra on Amazon Bedrock)
Agent output provenance — tracking trust and sources across multi-agent conversations
Proposal: GuardrailProvider protocol for tool call interception