Microsoft Agent Framework
Microsoft's open framework for building AI agents and multi-agent workflows in .NET and Python.
Graph-based workflows, middleware, OpenTelemetry and many model providers in one SDK.
Microsoft Agent Framework (MAF) is a code library for building AI agents and multi-agent workflows in Python and .NET. It suits development teams moving agents from prototype to production, especially on Azure or Microsoft Foundry.
What it is
You write agents in Python or C#, connect a model provider, and compose them into graph-based workflows with sequential, concurrent, handoff and group patterns. The README lists checkpointing, streaming, human-in-the-loop and time-travel, middleware, YAML declarative agents, Agent Skills, built-in OpenTelemetry observability, and a DevUI for debugging workflows.
Who it's for
It suits software teams who care about durability, governance and provider flexibility, and who already work in .NET or Python. It does not suit non-developers: there is no no-code builder in the README, and the quickstart authenticates through the Azure CLI.
Key features
Python and .NET parity
Consistent APIs across Python and C#/.NET packages.
Graph-based workflows
Sequential, concurrent, handoff and group collaboration patterns.
Checkpoints and time-travel
Workflows support checkpointing, streaming and human-in-the-loop.
OpenTelemetry built in
Distributed tracing and monitoring are part of the framework.
Declarative YAML agents
Define and version agents in YAML instead of code.
DevUI debugger
An interactive UI for developing and debugging agents and workflows.
Up and running in 3 steps
Our honest verdict
What's great
- MIT-licensed code with both Python (pip install agent-framework) and .NET (NuGet) packages
- Orchestration beyond a single prompt, with durable and restartable workflows
- Documented migration guides from Semantic Kernel and AutoGen
- Works with Microsoft Foundry, Azure OpenAI, OpenAI, the GitHub Copilot SDK and other providers
Where it falls short
- It is a developer framework, so you must write and host code; nothing here is point-and-click
- The README quickstart uses Microsoft Foundry and az login, so the smoothest path assumes an Azure account and paid model usage
- Fast-moving project with Python and .NET versioned separately and breaking changes flagged in commits
- Go support lives in a separate repository, microsoft/agent-framework-go
Pricing & license
The framework is free under the MIT licence. Costs come from the model provider or cloud services you connect, such as Microsoft Foundry or OpenAI, and from hosting your own app; the README shows no framework fee.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Is it free to use?
The framework is MIT-licensed and free. You pay whatever your chosen model provider or cloud charges for inference and hosting.
Which languages does it support?
Python and C#/.NET are in this repository. The README sends Go users to a separate agent-framework-go repository.
Does it replace Semantic Kernel or AutoGen?
The README links migration guides from both, so existing users have a documented path.
