Microsoft Agent Framework is Microsoft’s framework for building, orchestrating and deploying agents. It succeeds AutoGen and Semantic Kernel, combining the first’s multi-agent abstractions with the second’s enterprise features.
What it is for
- Individual agents that use models, tools and MCP servers
- Multi-agent graph workflows with sequential, concurrent, handoff and group-collaboration patterns
- Long-running executions with checkpointing, streaming and resumption
- Human-in-the-loop approval at sensitive steps
- Enterprise systems that require telemetry and strong typing
How it works
- The developer declares agents and connects nodes in a typed workflow
- Edges route data and activate each executor when inputs are ready
- Session state persists across turns and in checkpoints
- Supported providers include Azure OpenAI, OpenAI, Anthropic, Ollama, Microsoft Foundry and GitHub Copilot
- OpenTelemetry instrumentation for observability
License and maintenance
- Open source under the MIT license, maintained by Microsoft
- Stable 1.0 release for Python and .NET since April 2026
- AutoGen entered maintenance mode in October 2025; Microsoft recommends Agent Framework for new projects
Strengths
- Python and .NET parity, unusual among agent frameworks
- Enterprise features: session state, typing, filters and telemetry
- Documented migration path for AutoGen and Semantic Kernel projects
Limitations
- Fewer first-party integrations than the LangChain ecosystem
- In practice, closest to Azure and the Microsoft stack