← Back to Discover

Microsoft Agent Framework

Open-source Microsoft framework for creating AI agents and chaining them into multi-agent workflows, with Python and .NET versions, checkpointing, human-in-the-loop control and OpenTelemetry tracing.

Agent frameworks
Visit website ↗

What it does

Microsoft Agent Framework (MAF) is an open framework for teams that want to take agents from prototype to production. It is available for Python and C#/.NET with consistent APIs, and the repository points to a separate Go SDK. It works with providers including Microsoft Foundry, Azure OpenAI, OpenAI and the GitHub Copilot SDK, and the samples cover both local development and cloud deployment. Workflows are graphs that support sequential, concurrent, handoff and group collaboration patterns, with checkpointing, streaming, human-in-the-loop steps and time-travel. A middleware system handles request and response processing and exceptions, and built-in OpenTelemetry integration covers distributed tracing, monitoring and debugging. Other pieces include declarative agents defined in YAML, agent skills, Foundry-hosted agents that deploy with a couple of extra lines of code, a DevUI for testing and debugging, and experimental packages in AF Labs. The repository includes migration guides from Semantic Kernel and AutoGen. Python users install it with pip, and .NET users add the Microsoft.Agents.AI package. The repository notes that you are responsible for reviewing and testing the applications you build, and for the usage and costs of any third-party systems you connect.

Capabilities

  • Python and C#/.NET implementations with consistent APIs
  • Graph-based workflows with sequential, concurrent, handoff and group patterns
  • Checkpointing, streaming, human-in-the-loop control and time-travel in workflows
  • Built-in OpenTelemetry integration for tracing and monitoring
  • Declarative YAML agents, agent skills and a DevUI for testing
  • Support for several model providers, including Foundry, Azure OpenAI and OpenAI

When to use it

  • You are building agents that need to run in production with durability, observability and approval steps
  • You need orchestration beyond a single prompt or a stateless chat loop
  • You want to keep your choice of model provider open as requirements change
  • You are moving an existing Semantic Kernel or AutoGen project and want the migration guides