AG2
Open source multi-agent framework (formerly AutoGen)
AG2 is a Python framework for multi-agent systems that evolved from AutoGen and is maintained by volunteers from several organizations. Since version 1.0, multiple agents run via AG2 Network: a hub manages the registry, write-ahead log and audit trail, and agents communicate over typed channels (conversation, consulting, round-robin discussion, workflow with conditional handoffs). The classic API with ConversableAgent, Group Chat and Swarm moved to a separate package as AG2 Classic and is still maintained. AG2 is licensed under Apache-2.0; for organizations the website also lists the AG2 AgentOS runtime with access on request.
Good fit if
Focus: Developers who want fine control over conversation patterns between agents or continue existing AutoGen projects
- your project is about Discussions and reviews between agents, Multi-step workflows with handoffs or Research on multi-agent conversations
- someone on your team writes code
Probably not if
- nobody on your team writes code (1/5 for use without developers)
- there is little time for training (learning curve: steep)
Key figures
- Usable without developers
- 1/5
- Scalability
- 3/5
- Integrations
- 3/5
- Versioning
- 2/5
- Observability
- 3/5
Editorial rating on a scale from 1 to 5. Methodology
Fact sheet
- Category
- Developer
- Type
- Open source multi-agent framework (formerly AutoGen)
- Licence
- Open Source (Apache-2.0)
- Pricing
- Open SourceFramework without license fee (Apache-2.0). Costs arise for the models used. The website lists no prices for AG2 AgentOS; access is granted on request.
- Deployment
- Self-Hosted, Any
- Models
- OpenAI, Anthropic, Google Gemini, Ollama
- RAG
- Yes
- Multi-Agent
- Yes
- SSO, roles (RBAC)
- SSO no, RBAC no
- Latest version
- 1.1.2 (03.10.2026)
- GitHub stars
- 4,987
Key features
- AG2 Network with hub and typed channels
- Workflow channel with conditional handoffs
- Group Chat and Swarm in AG2 Classic
- Human in the loop
- Audit trail and replayable transcripts
Typical use cases
- Discussions and reviews between agents
- Multi-step workflows with handoffs
- Research on multi-agent conversations
- Migration of existing AutoGen applications
Market notes
Changes to AG2 by email
We let you know when the status, pricing or license of AG2 changes, together with the weekly market notes.
Implementation
Get AG2 implemented
Free for you. Partners only pay for requests they accept.
Alternatives
Similar systems
Similar in category, features or use cases.
- DeveloperAutoGenAutoGen is an open-source framework from Microsoft Research for multi-agent systems, known for group chats between agents and code execution. With the release of Microsoft Agent Framework 1.0 in April 2026, AutoGen entered maintenance mode: it only receives bug fixes and security updates, and new development happens in Agent Framework. Microsoft recommends Agent Framework for new projects. Independently of Microsoft, the community continues the fork AG2.View profile
- DeveloperAgentScopeAgentScope is a Python framework for agents and multi-agent systems from Alibaba. At its core is a ReAct agent with tool management for Python tools, MCP servers and skills, plus context compaction, RAG and long-term memory with backends such as Mem0 and ReMe. Multiple agents collaborate through workflows such as orchestrator-workers with handoffs, routing, debates and concurrent agents; in the Agent Service a leader agent spawns worker agents on demand (Agent Team), and remote agents are reachable via A2A. AgentScope Studio, tracing and evaluation are also included. The project is licensed under Apache-2.0.View profile
- DeveloperLangroidLangroid is a lightweight open source Python framework for applications built on LLM agents, developed by researchers from CMU and UW-Madison. Agents act as message transformers: a Task wraps an agent with instructions and orchestrates collaboration through hierarchical, recursive delegation to sub-tasks. For RAG there is the DocChatAgent with connections to vector databases such as Qdrant, Chroma, LanceDB or PGVector, and MCP tools can be integrated. Langroid is MIT licensed and works with commercial and local models.View profile
- DeveloperMicrosoft Agent FrameworkMicrosoft Agent Framework is Microsoft's open-source framework for agents and multi-agent workflows in Python and .NET. Version 1.0 shipped in April 2026 and is the unified successor of Semantic Kernel and AutoGen. A graph-based workflow engine supports branching, parallelism, checkpoints and human approvals; ready-made patterns exist for sequential, concurrent, handoff, group chat and Magentic-One orchestration. MCP and A2A are supported.View profile