Checked against official docs

Microsoft Agent Framework

Provider: Microsoft · Checked Oct 28, 2026

Microsoft's open-source developer toolkit for building AI agents and multi-step workflows in Python, C#, and Go. It's the next generation of Semantic Kernel and AutoGen.

Visit the official website

Pricing modelUnknown for now
Where it runsPython, Csharp, Go
Good forOperations
CheckedOct 28, 2026
What it can do
Developer SDKMulti agentWorkflow orchestrationHuman in the loopCheckpointsMCP supportSelf hosting

Pricing

Based on Microsoft's overview (checked Oct 28, 2026). Microsoft's site has current cost details. Model and cloud costs depend on what you connect and weren't on the page, so they're unknown here for now.

Getting set up

According to Microsoft Learn: install with 'pip install agent-framework' (Python) or 'dotnet add package Microsoft.Agents.AI.Foundry --prerelease' (C#), then connect a model provider such as Microsoft Foundry. Go is in public preview.

The plain-English overview

Does your developer work in Microsoft's world and need to build custom AI agents? Microsoft Agent Framework is a developer toolkit for building AI agents and the workflows that connect them. Here's what Microsoft Learn says (we checked on Oct 28, 2026).

Microsoft calls it the direct successor to two earlier projects, Semantic Kernel and AutoGen, built by the same teams. If you've heard of either, this is where they're headed. Microsoft has migration guides for both.

It brings together four main pieces:

  • Agents: AI agents that use a model, call tools and MCP servers (MCP is a standard way to plug AI into other apps), and respond. Microsoft lists support for Microsoft Foundry, Anthropic, Azure OpenAI, OpenAI, Ollama, and more.
  • Harness Agent: a ready-made agent for long, multi-step tasks, with planning and to-do tracking, file access and memory, tool approvals, and monitoring.
  • Workflows: step-by-step paths that connect agents and regular code in a set order.
  • Integrations: connections to model providers, tools, memory, evaluation services, and more.

The docs also cover human-in-the-loop steps, checkpoints so a workflow can pause and resume, background agents, and self-hosting.

Here's a piece of advice from Microsoft's own docs that we really like: if you can write a simple function to handle a task, do that instead of using an AI agent. Use an agent for open-ended work, and a workflow when the steps are well defined.

Who's it for? Developers. It's available for Python and C#, and Go is in public preview. Microsoft also reminds builders that they're responsible for testing what they build and for any costs from third-party models and services.

Pricing: Microsoft's site has current cost details. You pay for the models and cloud services you connect, and those prices aren't on this page, so they're unknown here for now.

Sources

Where each fact came from

A quick note on affiliate links

If you buy through an affiliate link on this site, we may earn a commission. It doesn't cost you anything extra. Anything sponsored gets a clear label, and we never sell our rankings. Right now, no affiliate programs are connected. Here's the full affiliate disclosure.