Microsoft .NET Agent Framework Reaches RC
Microsoft's .NET Agent Framework has reached Release Candidate status, providing stable APIs for building multi-agent workflows. The framework is designed for creating coordinated AI systems and supports interoperability between .NET and Python environments.
- This framework is the successor to Microsoft's earlier AI agent projects, AutoGen and Semantic Kernel, combining AutoGen's multi-agent orchestration capabilities with Semantic Kernel's enterprise-grade features. Microsoft has indicated that while AutoGen and Semantic Kernel will receive security patches, all new feature development will be focused on the new unified framework. - It provides a unified programming model for both .NET and Python, allowing developers to build and orchestrate agents in either language. The Release Candidate (RC) status signifies that the API is now stable ahead of the full General Availability release. - The framework is designed to integrate with various large language models (LLMs), not just those from Microsoft, supporting connections to services from OpenAI, Anthropic, and local models via Ollama. - A key feature is the use of graph-based workflows that allow developers to explicitly define how multiple agents and functions connect and interact. This enables complex patterns like sequential, concurrent, and handoff-based agentic systems. - For production environments, the framework includes built-in observability via OpenTelemetry for tracing agent conversations and performance. It also integrates with Azure AI Foundry for cloud deployment, which provides managed services for hosting and scaling agents. - To simplify debugging, a developer UI dashboard called "DevUI" is included, which provides a live visualization of the agent's internal reasoning steps, message exchanges, and tool usage. - The framework enables agents to maintain context across conversations through "Persistent Conversations," allowing the history of an agent's interactions to be saved and reloaded. It also supports features for long-running tasks and human-in-the-loop scenarios.
