Microsoft Agent Framework Samples: a hands-on path through building agents
Agent Framework Samples - showcasing ways in which agent framework can be utilized.
At a glance
- What is it?
- Microsoft Agent Framework Samples is an MIT-licensed collection of tutorials and runnable examples for the Microsoft Agent Framework, in both Python and .NET, covering foundations, tools, providers, RAG, multi-agent systems and workflows. It is a learning resource, not the framework itself.
- Who is it for?
- Work through Microsoft Agent Framework Samples if you are learning or evaluating the Microsoft Agent Framework and want runnable Python and .NET examples across providers, tools, RAG, multi-agent and workflows. Do not treat the examples as production code or expect them to run without cloud credentials, and check any sample against the framework's current docs since it tracks a moving API.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 82 days ago.
- What is it written in?
- Mainly Jupyter Notebook, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 18, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
What the samples repository offers
Microsoft Agent Framework Samples is a curated, hands-on guide to building agents with the Microsoft Agent Framework, provided by Microsoft alongside the framework itself. It is organized as numbered modules that move from beginner concepts to real cases, with runnable examples in both Python and .NET at nearly every step. The topics span agent foundations, creating a first agent, exploring providers and configurations, tools such as vision and code interpretation, provider patterns including MCP and agent-to-agent communication, RAG with file search, multi-agent orchestration, workflows, and evaluation and tracing. The audience is developers evaluating or learning the Microsoft Agent Framework who want working code to read and run rather than only reference documentation. It is explicitly a samples and tutorial repository, so its job is to teach the framework, not to be a library you depend on.
A dual Python and .NET curriculum
The structure is the substance here. The modules are ordered, from a beginners section that extends Microsoft's AI Agents for Beginners material, through building your first travel-planning agent, into provider deep dives and advanced multi-agent and workflow topics, ending with evaluation and tracing. What sets it apart from many sample sets is the consistent dual-language coverage: most topics ship both a .NET example and a Python notebook, so a team can follow the same concept in whichever runtime it uses. The provider examples are concrete, naming Azure OpenAI, GitHub Models, Microsoft Foundry and Foundry Local, which shows the framework working against different backends rather than one. That breadth makes the repository a map of what the framework can do as much as a set of exercises.
Getting the samples running
Because it is a samples repository rather than a package, using it means cloning it and running the examples for your runtime. The repository includes an Installation section and a devcontainer configuration, and the examples come as .NET projects and Python notebooks under the numbered module folders, with an environment example file for the keys and endpoints a given sample needs. The path is to set up the environment following the Installation guidance, choose a module, and open its .NET project or its Python notebook, supplying credentials for the provider that sample targets, such as GitHub Models or Azure OpenAI. The first real use is running the first-agent travel-planning example end to end, which exercises the framework's core loop before you move on to tools, RAG or multi-agent modules.
Where a samples repository stops being enough
The limitations are those of any teaching resource. These are examples, not production code, so they are meant to be read and adapted rather than deployed as-is, and they depend on the Microsoft Agent Framework and on Microsoft-adjacent providers such as Azure OpenAI, GitHub Models and Foundry, which shapes who they serve. Many samples need cloud credentials and, for some providers, paid services, so running everything is not free. As a fast-moving framework's companion, the samples track a moving target, and the last push was on 2026-06-28, so a given example can lag the framework's current API and should be checked against the framework's own documentation. None of this is a fault; it is what a samples repository is, and treating it as a curriculum rather than a dependency is the right posture.
The samples versus the framework docs or other tutorials
The alternatives are the Microsoft Agent Framework's own reference documentation, or third-party agent tutorials built on other frameworks. Reference docs are authoritative and current but abstract, telling you what each API does without a worked, runnable path through building something. Third-party tutorials may be engaging but usually cover one language and one stack and are not maintained by the framework's authors. This repository's difference is that it is the official, dual-language, end-to-end sample set, so the examples are meant to match the framework and to show the same concept in both Python and .NET. Use the reference docs to look up specifics and this repository to learn by running; for teams committed to the Microsoft Agent Framework, the official samples are the most direct way to get productive.
MIT license and maintenance
The repository is MIT-licensed, so you can reuse and adapt the sample code freely, which is exactly what a teaching resource should allow. The last push was on 2026-06-28, and the content is broad, from beginner modules to evaluation and tracing, with the standard open-project files such as a code of conduct, security and support docs. Because it accompanies a framework under active development, treat the samples as a snapshot: pin the framework version you learn against, and if an example does not run, check it against the framework's current documentation before assuming the sample is wrong. Start with the beginners and first-agent modules to confirm your environment and credentials work, then follow the numbered path into the topics you actually need.
Editorial conclusion
Work through Microsoft Agent Framework Samples if you are learning or evaluating the Microsoft Agent Framework and want runnable Python and .NET examples across providers, tools, RAG, multi-agent and workflows. Do not treat the examples as production code or expect them to run without cloud credentials, and check any sample against the framework's current docs since it tracks a moving API. Start by following the Installation guidance, then run the first-agent travel-planning example before moving through the numbered modules.
Frequently asked questions
What is Microsoft Agent Framework Samples?
It is an MIT-licensed official collection of tutorials and runnable examples for the Microsoft Agent Framework, in both Python and .NET, covering foundations, tools, providers, RAG, multi-agent systems, workflows and evaluation.
Is it the Agent Framework itself?
No. It is a samples and tutorial repository that teaches the Microsoft Agent Framework by example. You still use the framework and its providers, such as Azure OpenAI or GitHub Models, which the samples target.
How do I run the samples?
Clone the repository, follow its Installation section, and open a module's .NET project or Python notebook, supplying credentials for the provider that sample uses. Start with the beginners and first-agent modules.
Community notes