Microsoft .NET Agent Framework Reaches Release Candidate
Microsoft's Agent Framework for .NET has reached Release Candidate (RC) status. The framework is designed to support the creation of multi-agent workflows and is interoperable with both .NET and Python, providing a new tool for enterprise engineers building agentic systems.
- This framework is the official successor to Microsoft's earlier AI projects, AutoGen and Semantic Kernel, combining AutoGen's multi-agent orchestration with Semantic Kernel's enterprise features like telemetry and compliance hooks. - It introduces graph-based workflows, allowing developers to explicitly define and orchestrate complex interactions between multiple agents and functions, with support for sequential, concurrent, and handoff patterns. - The framework is built on open standards, including Agent-to-Agent (A2A) and Model Context Protocol (MCP), to ensure interoperability between agents built in different languages or by different vendors. - For observability, it integrates with OpenTelemetry, enabling developers to trace, monitor, and debug agent behavior in production environments. - It is designed to be multi-provider, offering support for models from Microsoft Foundry, Azure OpenAI, OpenAI, Anthropic, AWS Bedrock, and local models via Ollama. - The new framework provides a unified API for both Python and .NET, allowing teams with diverse language expertise to collaborate on building agentic systems. - A key feature for developers is an interactive DevUI for local development, which helps in testing and debugging agent workflows before deployment. - While the new Agent Framework is the focus for future development, Microsoft will continue to provide bug fixes and critical security patches for AutoGen.
