Skip to content
Get packs

Microsoft Agent Framework

A Microsoft framework for building and running AI agents and multi-agent workflows in Python and .NET.microsoft/agent-framework

GitHub Download python-1.24.0
Stars
14.5k
Forks
2.6k
Licence
MIT
Language
Python
Last update
4 days ago
Type
Library

What it is

Microsoft Agent Framework is a set of code libraries for building AI agents, which are programs that use language models to carry out tasks, and for connecting several agents into workflows. It supports Python and .NET with similar APIs, and includes features such as orchestration patterns, middleware, OpenTelemetry tracing, and a developer UI. Developers use it to move agent prototypes toward production with an approach that is not tied to a single model provider.

  • Create a named agent with instructions and run it against a Microsoft Foundry model, as shown in the quickstart
  • Build workflows where several agents run sequentially, concurrently, by handoff, or in group collaboration
  • Add checkpointing, streaming, or human-in-the-loop review to an agent workflow
  • Deploy an agent to Foundry-hosted infrastructure with a few lines of added code, according to the README
  • Define agents declaratively in YAML using the declarative-agents samples

Great for

  • Developers building agents or multi-agent workflows they plan to run in production
  • Teams that need orchestration beyond a single prompt or stateless chat loop
  • Python or C#/.NET developers who want one framework across both languages
  • Teams that want to switch model providers without rewriting their architecture
  • Developers who use Microsoft Foundry or Azure OpenAI

Not ideal for

  • People who only want a simple chatbot or a single prompt call; the framework is built for multi-step orchestration and may be more than needed
  • Non-developers, since setup requires writing code and configuring cloud credentials
  • Teams that need Go support from this repo; the README says Go lives in a separate repository, microsoft/agent-framework-go
Step-by-step guide: how to use Microsoft Agent FrameworkInstall, first run, examples, tips and FAQs
Microsoft Agent Framework alternativesMicrosoft Agent Framework vs Semantic KernelMicrosoft Agent Framework vs LangGraphMicrosoft Agent Framework vs CrewAI

How to get it

  1. 1

    Install with a package manager

    You'll need: Python (version not stated in the README) · pip · Azure CLI for the Python quickstart (az login) · Access to a Microsoft Foundry project for the quickstart · dotnet SDK for the .NET path (version not stated in the README)

    Install the Python package

    pip install agent-framework

    Add the .NET core package

    dotnet add package Microsoft.Agents.AI
  2. 2

    Run it

    • Install the Python or .NET package using the commands above
    • Sign in with Azure CLI using az login, or configure DefaultAzureCredential for .NET
    • Set FOUNDRY_PROJECT_ENDPOINT (and optionally FOUNDRY_MODEL) as environment variables
    • Run the Basic Agent Python quickstart, which asks the agent to write a haiku
  3. 3

    Get the most out of it

    • Use the samples folders for providers, middleware, workflows, and observability, which the README links to by language
    • Enable the built-in OpenTelemetry integration to trace and debug agent runs
    • Try the DevUI to test and debug workflows interactively
    • Read the MS Learn documentation and migration guides if you are coming from Semantic Kernel or AutoGen

Downloads

Source code (.zip)main branch · straight from GitHubRelease python-1.24.08 files incl. agent_framework-1.24.0-py3-none-any.whlOfficial website & docsaka.ms/agent-framework

Strengths

  • Supports both Python and .NET with documented samples for each
  • Includes built-in OpenTelemetry observability and middleware for custom request handling
  • Offers graph-based orchestration patterns and provider flexibility in one framework

Trade-offs

  • The README is truncated and several features, such as time-travel and the Labs packages, are described only briefly, so details need checking in the docs
  • Quickstarts assume Azure and Microsoft Foundry access, which may be a barrier outside that ecosystem
  • Some components are marked experimental (AF Labs), so stability varies by package

How it compares

ProjectStarsTypeLevelUpdatedBest for
agent-framework (this)14.5kLibraryintermediate4 days agoDevelopers building agents or multi-agent workflows they plan to run in production
langgraph44.3kLibraryintermediate4 days agoPython developers building multi-step AI agents that must survive failures or long waits
crewAI60.4kLibraryintermediate4 days agoPython developers prototyping multi-agent workflows
semantic-kernel28.7kLibraryintermediate4 days agoDevelopers already using Azure OpenAI or OpenAI who want a structured SDK for agents
ruflo80.6kDev toolintermediate4 days agoDevelopers already using Claude Code or Codex who want multiple agents working together
PocketFlow11.3kLibraryintermediate3 months agoDevelopers who want a small, readable LLM framework they can inspect in full

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.