Microsoft has unveiled the public preview of the Agent Framework, merging its previous Semantic Kernel and AutoGen offerings into a cohesive open-source SDK designed to facilitate the development of AI agent applications. This framework is built on .NET 9+ and offers robust support for building multi-agent AI systems.
The Agent Framework provides a versatile approach to workflow and agent orchestration, supporting predefined business processes and dynamic, LLM-driven interactions. Developers can choose from various orchestration models, including sequential, concurrent, group chat, hand-off, and magnetic workflows. With OpenAPI integration, the framework ensures seamless service connections, while its cross-platform portability allows for both cloud-based and on-premises deployment.
Notably, the Agent Framework supports cloud LLMs alongside local NPU-optimized SLMs, enhancing its adaptability across different use cases. Developers benefit from abstracted client interfaces that standardize agent development, as well as integration with the Model Context Protocol and emerging industry standards.
The framework is also designed for local experimentation with the capacity for broad deployment via Azure AI Foundry. It boosts observability through OpenTelemetry contributions and offers the Voice Live API for real-time speech interactions. Furthermore, responsible AI features such as task adherence and prompt shields ensure ethical AI operations.
Organizations like KPMG, Commerzbank, and Citrix have already started utilizing the framework to construct production-ready AI solutions. Meanwhile, Microsoft assures developers that the existing Semantic Kernel will continue to receive critical bug and security support for at least a year following the Agent Framework’s general availability. Dedicated migration guides assist developers in transitioning from Semantic Kernel and AutoGen to the new framework in both .NET and Python.
In summary, the Microsoft Agent Framework represents a significant step forward in AI development, offering powerful tools and integrations that promote efficient creation of intelligent, versatile agent applications.

Collections

Microsoft released a public preview of its Agent Framework, an open-source SDK that unifies Semantic Kernel and AutoGen for building multi-agent AI systems on .NET 9+. The framework supports multiple orchestration models (sequential, concurrent, group chat, hand-off, magnetic), integrates with OpenAPI and Model Context Protocol, and works with both cloud LLMs and local NPU-optimized SLMs. It includes responsible AI features, OpenTelemetry observability, Voice Live API for real-time speech, and can be deployed locally or via Azure AI Foundry. Companies like KPMG, Commerzbank, and Citrix are already using it for production solutions, while Semantic Kernel will receive support for at least one year post-GA with migration guides available for .NET and Python.

Microsoft Agent Framework: A Unified SDK for Multi-Agent AI Systems