Microsoft Agent Framework (github.com)

Microsoft released a multi-language Agent Framework for building, orchestrating, and deploying AI agents with first-class support for both Python and .NET. Install via pip (agent-framework) or the Microsoft.Agents.AI NuGet package; the framework exposes consistent APIs across languages and includes a DevUI for interactive development, testing and debugging. It covers everything from simple chat agents to complex multi‑agent graph workflows and ships samples and quickstarts (including Azure OpenAI/Responses client examples that use AzureCliCredential to create agents). Technically, the framework emphasizes production-ready orchestration: graph-based workflows compose agents and deterministic functions with streaming, checkpointing, human-in-the-loop integration and “time‑travel” (replayable) execution. Built-in middleware hooks enable custom request/response pipelines and exception handling, while OpenTelemetry integration provides distributed tracing and observability for monitoring and debugging. AF Labs exposes experimental tooling for benchmarking, reinforcement learning and research, and the framework supports multiple LLM providers out of the box. For the AI/ML community this accelerates safe, reproducible multi-agent systems and enterprise integrations (notably Azure), simplifies troubleshooting at scale, and standardizes extensible agent patterns for both research and production.