Microsoft Agent Framework Hands-On Samples
Learn to build, orchestrate, and debug Microsoft Agent Framework agents with Python and .NET examples.
What does this agent do, and when should you use it?
This is a tutorial and sample repository for Microsoft Agent Framework, not a single deployable agent product. Its progression spans fundamentals, travel-planning agents, tools, RAG, providers, workflows, evaluation and tracing, and case studies. Samples are available in both Python and .NET, with examples using Azure OpenAI, GitHub Models, Microsoft Foundry, or Foundry Local. The deliverables are tutorials, notebooks, .NET code samples, and DevUI examples that users run locally after configuring runtimes, credentials, and selected services. It fits developers evaluating or learning Microsoft Agent Framework integrations; a production-ready agent still requires selecting a sample and adding deployment and application logic.
The repository uses chapter README files, Python notebooks, and .NET code samples to demonstrate travel-planning agents, model-provider configurations, and Vision, Code Interpreter, Bing Grounding, File Search, and custom-tool integration. Its RAG examples use File Search for knowledge-enhanced flows; 05.Providers includes an MCP example with Microsoft Learn; and 07.Workflow demonstrates Basic Workflow, Sequential, Concurrent, and Conditional workflows. 08.EvaluationAndTracing includes a GitHub Models Workflow DevUI plus single-agent and multi-agent DevUI examples for Microsoft Foundry, GitHub Models, and Foundry Local. Users install dependencies, create a root .env containing endpoints, model deployment names, and tokens, then run the chosen code sample or notebook.
- A Python developer wants to build a first travel-planning agent and start with a GitHub Models example to understand the basic flow.
- A .NET developer needs to examine sample configurations for Azure OpenAI, GitHub Models, Microsoft Foundry, and Foundry Local.
- A team building knowledge-assisted question answering wants to use the File Search RAG sample as a reference for adding file search to an agent flow.
- An engineer who must orchestrate agents sequentially, concurrently, or through conditional branches can study the matching examples in 07.Workflow.
- A developer investigating agent behavior or workflow execution can try the DevUI examples in 08.EvaluationAndTracing.
What are this agent's strengths and limitations?
- Provides both Python and .NET samples for learning comparable topics across two development stacks.
- Covers a defined path from basic agents through RAG, MCP, multi-agent workflows, evaluation, and tracing.
- Offers concrete sample entry points for Azure OpenAI, GitHub Models, Microsoft Foundry, and Foundry Local.
- Workflow coverage explicitly includes basic, sequential, concurrent, and conditional orchestration, alongside DevUI-oriented debugging examples.
- It is a collection of tutorials and samples, not a unified agent application with a direct deployment path.
- Many samples depend on external services, endpoints, model deployments, and authentication, including Azure OpenAI, Microsoft Foundry, GitHub Models, Azure CLI, and Azure Developer CLI.
- The documentation does not specify one top-level command to run all samples or complete deployment instructions for each sample.
- Some capabilities are provider-bound, including Bing Grounding, Microsoft Foundry Agent Service, and Foundry Local, so portability can require configuration or implementation changes.
How do you install or deploy this agent?
The documented runtimes are Python 3.10+ and .NET 9.0+. For the released version, run pip install agent-framework --pre for Python, or run dotnet add package Microsoft.Agents.AI in a .NET project. For build-from-source features, run pip install -r ./Installation/requirement.txt -U for Python; for .NET, run git clone https://github.com/microsoft/agent-framework.git, change to agent-framework/dotnet, then run dotnet build agent-framework-dotnet.slnx.
How do you use this agent?
Create a root .env and supply configuration for the selected sample: GitHub Models uses GITHUB_TOKEN, GITHUB_ENDPOINT, and GITHUB_MODEL_ID; Azure OpenAI uses AZURE_OPENAI_ENDPOINT and AZURE_OPENAI_CHAT_DEPLOYMENT_NAME; Microsoft Foundry uses AZURE_AI_PROJECT_ENDPOINT and AZURE_AI_MODEL_DEPLOYMENT_NAME. Set FOUNDRYLOCAL_ENDPOINT, FOUNDRYLOCAL_MODEL_DEPLOYMENT_NAME, BING_CONNECTION_ID, BING_CONNECTION_NAME, and OTEL_EXPORTER_OTLP_ENDPOINT when the selected sample needs them. Then run the linked Python notebook or .NET code sample; no single top-level launch or deployment command is documented.
How does this agent compare with similar options?
The repository presents separate sample paths for Azure OpenAI, GitHub Models, Microsoft Foundry, and Foundry Local; it does not provide performance, cost, or feature-comparison conclusions between them.