MicrosoftAutoGen is a framework for creating multi-agent AI applications that can act autonomously or work alongside humans. It is in maintenance mode: it receives no new features, is community managed, and Microsoft Agent Framework is its designated successor, with a migration guide for existing users.
The framework uses a layered design. The Core API implements message passing, event-driven agents, and local and distributed runtimes, with cross-language support for .NET and Python. The AgentChat API builds on Core with an opinionated interface for rapid prototyping and common multi-agent patterns such as two-agent chat and group chats. The Extensions API supplies LLM clients (OpenAI, Azure OpenAI, and others) and capabilities such as code execution.
Agents can use MCP servers through McpWorkbench, and AgentTool wraps one agent as a tool for another to form basic orchestration setups. Two developer tools accompany the framework: AutoGen Studio, a no-code GUI for prototyping multi-agent workflows (not intended as a production app), and AutoGen Bench for evaluating agent performance. Magentic-One, a multi-agent team for web browsing, code execution, and file handling, is built on the AgentChat and Extensions APIs. Python 3.10 or later is required.
Features
- Core API: message passing, event-driven agents, and local or distributed runtimes
- AgentChat API: higher-level interface for two-agent chat, group chat, and other multi-agent patterns
- Extensions API: first- and third-party extensions such as model clients and code executors
- MCP support: McpWorkbench connects agents to MCP servers such as Playwright
- Agent-as-tool orchestration: AgentTool lets one agent call specialist agents as tools
- AutoGen Studio: no-code GUI for prototyping multi-agent workflows
- AutoGen Bench: benchmarking suite for evaluating agent performance
- Cross-language runtime: Python and .NET packages for Core and gRPC runtime gateway
- Maintenance mode: bug fixes, security patches, and documentation only; new work is directed to Microsoft Agent Framework
Integrates with
Integrated by
Agent Governance Toolkit
Policy enforcement, zero-trust identity, execution sandboxing, and audit logging for autonomous AI agents across frameworks and languages
Forkd
Agent sandboxes need VM-grade isolation without booting a fresh VM every time.
Future AGI
OpenTelemetry-based tracing and evaluation instrumentation for voice-AI applications.
Gateway
Open-source AI gateway routing to 1,600+ LLMs through one API with retries, fallbacks, and guardrails
Alternatives
Langroid
Python framework for building LLM applications with collaborating multi-agent message passing
-
Microsoft Agent Framework
Successor SDK merging AutoGen and Semantic Kernel into one agent-building library.
Ruflo
Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI sys