Microsoft has announced the Release Candidate (RC) of its Agent Framework, a robust toolkit designed for developers working with .NET and Python to build intelligent AI agents. As the framework reaches RC status, it is deemed stable and feature-complete, paving the way for its general availability.
The Microsoft Agent Framework merges previous initiatives like Semantic Kernel and AutoGen to form a comprehensive software development kit (SDK). This SDK is tailored to assist in the creation, orchestration, and deployment of AI agents, integrating structured abstractions and tools within the Microsoft ecosystem.
Key features of the framework include simple agent creation requiring minimal code, type-safe function tools, and graph-based workflows that support both sequential and concurrent agent orchestration. It offers human-in-the-loop support, enabling more interactive AI applications, along with streaming responses, checkpointing, and Managed Communication Protocol (MCP) interoperability.
For developers transitioning from Semantic Kernel and AutoGen, Microsoft is preparing detailed migration guides. Additionally, current packages are available as pre-release versions on NuGet and PyPI, with extensive documentation and examples accessible through the official docs site and GitHub repository. This framework is set to simplify the complexities involved in agentic AI development, making it more accessible for developers within the Microsoft ecosystem.

Collections

Microsoft has released the Release Candidate of its Agent Framework, a unified SDK for building AI agents in .NET and Python. It merges Semantic Kernel and AutoGen into a single toolkit offering simple agent creation, type-safe function tools, graph-based workflows with sequential and concurrent orchestration, human-in-the-loop support, streaming responses, checkpointing, and MCP interoperability. Pre-release packages are available on NuGet and PyPI, with migration guides planned for developers coming from Semantic Kernel and AutoGen.

Microsoft Agent Framework Release Candidate Streamlines AI Development in .NET and Python