AutoGen and AG2
AutoGen is a Microsoft framework for multi-agent applications. The repository has since been archived; Microsoft points to Microsoft Agent Framework as its successor. AG2 is an independently continued project that grew out of AutoGen.
- Project
- AutoGen
- Publisher
- Microsoft
- Repository
- github.com/microsoft/autogen
- Documentation
- microsoft.github.io/autogen/
- Languages
- Python
- Category
- AI agent frameworks
What is AutoGen?
AutoGen is a Python framework by Microsoft for applications in which several AI agents work autonomously or alongside people. It includes the AgentChat layer for agents and teams, extensions for model providers and MCP servers, and AutoGen Studio as a no-code interface.
As of October 2026, the repository is archived. According to its README, Microsoft no longer provides bug fixes, security updates or new features and recommends Microsoft Agent Framework to new users; a migration guide is available for existing projects.
What is AG2?
AG2 (formerly AutoGen) is developed by a group of volunteer maintainers from several organisations and released under the Apache 2.0 licence. With version 1.0, AG2 introduced a new agent model: the package is called ag2, and collaboration between agents runs through a “network” of a hub and typed channels. The hub keeps the registry, a write-ahead log and an audit trail.
The classic variant derived from AutoGen, with ConversableAgent and GroupChat, continues as AG2 Classic in a separate repository and is still maintained. AG2 1.0 is not a drop-in replacement for Classic; imports, the agent model and orchestration have changed.
Core capabilities of AG2
- Agents and tools
- Python functions become tools via a decorator; AG2 runs the complete tool-calling loop.
- Human-in-the-loop
- A tool can pause the run and ask a person; a dedicated hook defines how the answer is obtained.
- Channels for several agents
- Two-party conversation, strict question-and-answer consulting, round-robin discussion and declarative workflows with conditional handoffs.
- Knowledge and compaction
- A knowledge store that persists across runs and strategies that summarise long histories instead of cutting them off.
Using multi-agent frameworks effectively in an AI-first business model
The history of AutoGen shows that choosing a framework is also an operating decision: who maintains the code, how are security issues fixed, and what does a migration cost? These questions belong to planning an agent just as much as outcome and mandate.
Keep business logic, approval rules and evaluation cases as independent of the framework as possible. Then an agent can be moved to a different technical foundation without losing accountability or quality criteria.
From tool to outcome
In AI-first coaching you clarify which outcomes AI agents should be accountable for in your go-to-market and within which framework. REVAGO then designs, builds and runs your system. You pay for results – not for activity.
Frequently asked questions
Is AutoGen still being developed?
The microsoft/autogen repository is archived. Microsoft no longer provides bug fixes, security updates or new features for it and recommends Microsoft Agent Framework as its successor, including a migration guide.
What is the difference between AutoGen and AG2?
AG2 is an independent open-source project that grew out of AutoGen and is continued by volunteer maintainers. AG2 1.0 introduces a new agent model with a hub and channels; the classic AutoGen-like variant is called AG2 Classic and lives in its own repository.
Does existing AutoGen code run on AG2 1.0?
Not as is. According to AG2, code using import autogen, ConversableAgent or GroupChat belongs to AG2 Classic (pip install ag2-classic). A dedicated migration guide covers the move to AG2 1.0.
Further reading in the wiki
A neutral overview, not a recommendation of individual providers. REVAGO is not affiliated with the listed projects. Descriptions are based on the projects’ public information; features and licences may change.