Microsoft Agent Framework

Microsoft merged the AutoGen research lineage and the production-focused Semantic Kernel into the Agent Framework, its unified open-source foundation for AI agents and multi-agent workflows across .NET and Python. Workflows are graph-based with sequential, concurrent, handoff, and collaboration orchestration patterns, and the framework layers on a middleware system for custom pipelines, OpenTelemetry-based observability, declarative agent definitions in YAML, and a DevUI for local testing and debugging. Multiple LLM providers are supported, with first-class paths to Azure AI Foundry including hosted agent deployment. Installation is pip install agent-framework for Python or the Microsoft.Agents.AI packages on NuGet for C#, and official migration guides walk existing AutoGen and Semantic Kernel codebases over. The whole framework is MIT licensed, which is unusual among the large-vendor agent stacks, and development is brisk: about 12.5k stars, more than 2,600 commits, and extensive sample sets running from getting-started material through end-to-end applications, documented at learn.microsoft.com/agent-framework.