Microsoft Agent Framework is Microsoft’s open-source framework for building, orchestrating, and deploying production-grade AI agents and multi-agent workflows. It is the convergence point of the Semantic Kernel and AutoGen lines, redesigned for teams taking agents from prototype to real systems. The project is under intense active development on GitHub, where it has earned more than 14,000 stars, with commits landing through late 2026.
The framework provides first-class SDKs for Python and .NET, with a Go SDK alongside, so enterprise teams can build agents in the languages they already run. Its centerpiece is graph-based workflow orchestration: compose agents and functions into sequential, concurrent, handoff, and group-collaboration patterns, with checkpointing and durability so long-running work can pause, resume, and survive restarts. Human-in-the-loop controls, observability, and governance hooks address the requirements that appear as soon as agents touch real business processes. Provider flexibility is a design goal: the same architecture works across Microsoft Foundry, Azure OpenAI, OpenAI, and other model services, including the GitHub Copilot SDK, and hosting patterns cover both local development and cloud deployment. Extensive samples demonstrate everything from single agents with tools to complex multi-agent systems.
Microsoft Agent Framework is free and open source under the MIT license. You pay only for the model APIs and infrastructure you choose to run it on, such as Azure OpenAI usage, and there is no framework subscription or per-agent fee.
The framework targets engineering teams, especially in Microsoft-centric organizations, that need agents to survive contact with production: orchestrated, observable, and governable. If you outgrew a single prompt loop or are migrating from Semantic Kernel or AutoGen, this is Microsoft’s designated path forward.