Microsoft Agent Framework Documentation
by Microsoft
The successor to Semantic Kernel and AutoGen, with migration guides from both.
Overview
Microsoft Agent Framework is an MIT-licensed multi-language framework (13k-plus GitHub stars, 2.2k forks) for production agents and multi-agent workflows, built by the teams behind Semantic Kernel and AutoGen and positioned explicitly as the successor to both: AutoGen's simple agent abstractions plus Semantic Kernel's session-based state management, type safety, middleware and telemetry, with graph-based workflows added for explicit orchestration. The Learn documentation opens with a seven-step Get Started path covering your first agent and tools, multi-turn conversations, memory and persistence, workflows, the agent harness, and hosting. Concepts covers agents, workflows, the harness, conversations and memory, and middleware. Agent Capabilities covers tools, agent skills, RAG, security and background agents; Workflow Capabilities covers agents in workflows, human-in-the-loop, checkpoints and resuming, and orchestrations. Integrations and Hosting are organised both by provider and by component, with dedicated pages for self-hosting and for Azure Functions durable hosting. The Support section carries an FAQ plus first-party migration guides from AutoGen and from Semantic Kernel. Four building blocks are named in the overview: Agents, which call tools and MCP servers across Microsoft Foundry, Anthropic, Azure OpenAI, OpenAI and Ollama; the Harness Agent, an opinionated agent with planning and todo tracking, context compaction, file access and memory, don't-ask-again tool approval and observability for long multi-step tasks; Workflows; and Integrations. Packages are agent-framework on PyPI and Microsoft.Agents.AI on NuGet, with OpenTelemetry tracing and a DevUI included. Two caveats the docs state plainly: the Go SDK is in public preview and still lacks declarative agents, RAG, CodeAct and functional workflows, and the overview advises that if you can write a plain function to handle a task you should do that instead of using an agent.
At a Glance
- Topic
- Agentic
- Level
- Intermediate
- Format
- Documentation
- Cost
- Free
- Duration
- ~4-6 hours through the seven-step Get Started path; reference docs are ongoing
- Provider
- Microsoft
- Hands-on
- Yes — code/exercises
- Certificate
- None
What You’ll Learn
- ✓Build a first agent with tools in C#, Python or Go
- ✓Choose between an autonomous agent and an explicit graph workflow
- ✓Add multi-turn conversation state and persistent memory to an agent
- ✓Connect hosted MCP servers as tools an agent can call
- ✓Insert human-in-the-loop approval steps and resume runs from checkpoints
- ✓Instrument agents with OpenTelemetry and debug them through DevUI
- ✓Migrate an existing AutoGen or Semantic Kernel project across
- ✓Host agents self-managed or on Azure Functions durable execution
Highlights
- •The same API surface across .NET and Python, with Go in public preview
- •First-party migration guides from both AutoGen and Semantic Kernel
- •The Harness Agent ships planning, context compaction and tool approval out of the box
- •OpenTelemetry tracing and a DevUI are included rather than bolted on later
- •The overview opens by telling you when not to use an agent at all
Who It’s For
Best For
- ✓Enterprise .NET teams adding agents to existing production services
- ✓Existing Semantic Kernel or AutoGen users planning a migration
- ✓Teams that need explicit orchestration control rather than free-running agents
- ✓Developers deploying to Microsoft Foundry or Azure Functions
Prerequisites
- •Working knowledge of C#, Python or Go and its package manager
- •An LLM endpoint such as Azure OpenAI, OpenAI, Anthropic or Ollama
- •Azure CLI credentials for the Foundry-based quickstart samples
FAQ
What is Microsoft Agent Framework Documentation?
Microsoft Learn documentation for Agent Framework, the MIT-licensed successor to Semantic Kernel and AutoGen for building agents and multi-agent workflows in .NET, Python and Go. It covers a seven-step getting-started path, agent and workflow concepts, tools and hosted MCP servers, memory and middleware, human-in-the-loop, checkpointing and resuming, orchestration patterns, integrations by provider and component, hosting options, and migration guides from both predecessors.
Is Microsoft Agent Framework Documentation free?
Microsoft Agent Framework Documentation is free to access.
What level is Microsoft Agent Framework Documentation for?
Microsoft Agent Framework Documentation is aimed at a intermediate audience. Recommended background: Working knowledge of C#, Python or Go and its package manager, An LLM endpoint such as Azure OpenAI, OpenAI, Anthropic or Ollama, Azure CLI credentials for the Foundry-based quickstart samples.
How long does Microsoft Agent Framework Documentation take?
Expect roughly ~4-6 hours through the seven-step Get Started path; reference docs are ongoing. Most learners work through it at their own pace.
What will I learn from Microsoft Agent Framework Documentation?
You'll learn: Build a first agent with tools in C#, Python or Go; Choose between an autonomous agent and an explicit graph workflow; Add multi-turn conversation state and persistent memory to an agent; Connect hosted MCP servers as tools an agent can call; Insert human-in-the-loop approval steps and resume runs from checkpoints; Instrument agents with OpenTelemetry and debug them through DevUI; Migrate an existing AutoGen or Semantic Kernel project across; Host agents self-managed or on Azure Functions durable execution.
Topics
Sources
This page was written from 3 sources, 1 on domains other than learn.microsoft.com.