Microsoft Agent Framework
Open-source Microsoft framework for AI agents and multi-agent workflows in Python, .NET and Go, the successor to AutoGen and Semantic Kernel.
What it does
With Microsoft Agent Framework you build AI agents that use a language model, call tools and MCP servers, and keep conversations going across sessions. Workflows connect several agents and functions along defined paths – sequential, concurrent or with handoffs. Middleware, OpenTelemetry tracing and checkpoints for long-running processes with human approval are built in. Connectors exist for Microsoft Foundry, Azure OpenAI, OpenAI, Anthropic, Amazon Bedrock, Google Gemini and Ollama, among others. Version 1.0 for Python and .NET has been available since April 2026.
Who it suits
Development teams that want to run agents in their own code, especially in Microsoft and Azure environments. If you currently use AutoGen or Semantic Kernel, official migration guides are available.
Data protection for businesses
The framework is a library you run on your own infrastructure. Where inputs go, whether they are used for training and which data processing agreement applies depend on the model provider you choose. Microsoft does not mention a separate data processing agreement or business plans for the framework itself. Microsoft points out that you must check yourself whether data leaves your compliance and regional boundaries via third-party services.
Limits
Not usable without programming skills. Some parts such as DevUI, skills and the agent harness are still in preview, as is the Go version. You have to implement safeguards such as content filters yourself.