Buying snapshot
Scroll sideways to see all columns. With a keyboard, focus this table and use the arrow keys.
| Decision point | What Microsoft states |
|---|---|
License | Code in the repository under the MIT License; documentation under Creative Commons Attribution 4.0 International |
How you get it | pip install of autogen-agentchat and autogen-ext[openai]; requires Python 3.10 or later |
AutoGen Studio | No-code GUI installed as the autogenstudio pip package; for prototyping, not meant to be a production-ready app |
Project status | Archived and no longer maintained by Microsoft; now community-managed |
Support | Microsoft provides no bug fixes, security fixes, new features or support; community help on Discord and GitHub Discussions, where responses may be limited |
Choosing what to install
AgentChat. A simpler, opinionated API for rapid prototyping, built on the Core API. It supports common patterns such as two-agent chat and group chats.
Core API. Message passing, event-driven agents, and a local or distributed runtime, with support for .NET and Python. Choose it when you need lower-level control.
AutoGen Studio. Prototype and run multi-agent workflows without writing code. For a deployed app, Microsoft tells developers to build on the framework and add their own authentication and security.
How the cost is built
The MIT License covers the code, so the framework itself carries no license bill. What you pay for is the model behind your agents and the machines that run them. The Extensions API ships LLM clients for OpenAI and Azure OpenAI, and the quickstart samples call the OpenAI API with OpenAI's GPT models (gpt-4.1 in the code), which needs an OpenAI account and API key. Those model charges are set by the provider you pick, not by AutoGen.
What drives the bill
Expert agents. In the multi-agent sample, a general assistant uses a math expert and a chemistry expert as tools when needed, and each expert is an agent built on the model client. Each expert the assistant calls adds that expert's model use to the task.
Runtime. The Core API offers local and distributed runtimes. A distributed runtime means more hosts to run and pay for.
Costs that are easy to miss
Security work is yours. Vulnerabilities in AutoGen and its dependencies may stay unaddressed. Microsoft says to assess this risk carefully, especially with sensitive data, and advises against production use.
No new features. As AutoGen is in maintenance mode, contributions are limited to bug fixes, security patches and documentation improvements. Feature work goes to Microsoft Agent Framework, so any feature you need in AutoGen is yours to build and keep.
MCP servers. The web browsing sample uses the Playwright MCP server, installed separately with npm. Microsoft warns to connect only to trusted MCP servers, as they may execute commands in your local environment or expose sensitive information.
Migration effort. Teams upgrading from AutoGen v0.2 follow a migration guide to update code and configurations. Teams leaving AutoGen use the AutoGen → Microsoft Agent Framework migration guide.
Our take
If you need agents for research, experiments or learning, AutoGen costs only the model and hosting you choose. If you plan a production system, the free license does not buy maintenance or a supported path, so budget for Microsoft Agent Framework instead.
Alternatives
Microsoft Agent Framework. Microsoft calls it the enterprise-ready successor to AutoGen and tells new users to start there. Version 1.0 is a production-ready release with stable APIs and a commitment to long-term support, with multi-provider model support and interoperability via A2A and MCP.