Microsoft Agent Framework

Microsoft Agent Framework is an open source development environment for building generative artificial intelligence agents and multi-agent systems. This SDK unifies two previous Microsoft projects, Semantic Kernel and AutoGen, into a platform that integrates experimental capabilities and enterprise functionalities.

The framework enables creating agents that reason about objectives, invoke tools, collaborate with each other and adapt dynamically. It includes support for orchestration patterns such as sequential flows, concurrent processing, group collaboration and task handoff between agents. It also implements orchestration through manager agents that coordinate tasks between specialised agents and humans.

Microsoft Agent Framework is extensible and modular. Agents can be configured declaratively using YAML or JSON, and developed in Python or .NET, with SDKs available on PyPI and NuGet. They can also connect to services such as Azure AI Foundry, Microsoft Graph, Redis or any OpenAPI-compatible API. Support for standards such as MCP (Model Context Protocol) and A2A (Agent-to-Agent) provides interoperability across environments.

The SDK includes observability capabilities with OpenTelemetry, integration with Azure Monitor, authentication via Entra ID and compatibility with CI/CD pipelines. Agents can run in containers, locally or across multiple clouds, enabling deployment in regulated and high-availability environments.

The framework provides implementation examples requiring minimal code configurations. Its integration with Visual Studio Code and open approach position it as a tool for developing applications based on intelligent agents.