Microsoft Agent Framework Demos

A progressive series of .NET console applications demonstrating key concepts of the Microsoft Agent Framework. Each demo builds on the previous one, introducing new capabilities step-by-step.

What You'll Learn

• Setting up AI agents with streaming responses
• Managing conversation sessions
• Adding function calling (tools) to extend agent capabilities
• Implementing middleware for content filtering
• Building multi-agent workflows

Prerequisites

• .NET 10 SDK
• An Azure OpenAI resource (for demos 2 & 3) OR GitHub Models API access (for demo 1)
• Azure CLI (for authentication): az login

Project Structure

agent-framework-demos/
├── source/
│   ├── 1-HelloAgent/          # Basic agent with streaming
│   ├── 2-Extending/            # Tools + middleware
│   └── 3-Workflow/             # Multi-agent workflows
└── agent-framework-demos.sln   # Solution file

Demos Overview

1. HelloAgent - Your First Agent

Concepts: Agent setup, streaming responses, session management, interactive REPL

A frozen yogurt recommendation chatbot that demonstrates:

• Creating an AI agent with custom instructions
• Streaming responses token-by-token
• Maintaining conversation context with AgentSession
• Building a simple command-line interface

Configuration: Uses GitHub Models (OpenAI) by default

2. Extending - Tools & Middleware

Concepts: Function calling, middleware pipeline, content filtering

Extends the froyo agent with:

• Tools (Function Calling): Calls a ListFlavors() function to get real-time inventory
• Middleware: Filters forbidden words from input/output using a custom pipeline
• Local functions for helper logic

Configuration: Uses Azure OpenAI with Azure CLI authentication

3. Workflow - Multi-Agent Orchestration

Concepts: Agentic workflows, agent composition, order processing pipeline

A multi-step order processing system with:

• Order Intake Agent: Validates orders against business rules
• Froyo Recommender Agent: Product recommendations
• Demonstrates how agents can work together in a workflow

Configuration: Uses Azure OpenAI with Azure CLI authentication

Getting Started

1. Clone the Repository

git clone https://github.com/mcollier/agent-framework-demos.git
cd agent-framework-demos

2. Configure Settings

Each demo has an appsettings.json file. Copy it to appsettings.Development.json and add your credentials:

Demo 1 (GitHub Models)

{
  "OpenAI": {
    "Endpoint": "https://models.github.ai/inference",
    "DeploymentName": "openai/gpt-5-chat",
    "Key": "YOUR_GITHUB_PAT_HERE"
  }
}

Demos 2 & 3 (Azure OpenAI)

{
  "AzureOpenAI": {
    "Endpoint": "https://YOUR_RESOURCE.openai.azure.com/",
    "DeploymentName": "gpt-4o-mini"
  }
}

Note: appsettings.Development.json is git-ignored to keep secrets out of source control.

3. Authenticate with Azure (for demos 2 & 3)

az login

The demos use AzureCliCredential for passwordless authentication.

4. Run the Demos

Navigate to each demo folder and run:

cd source/1-HelloAgent
dotnet run

# Then try the others
cd ../2-Extending
dotnet run

cd ../3-Workflow
dotnet run

Type exit or press Enter on an empty line to quit.

Key Concepts Demonstrated

Agents

An AI Agent combines an LLM with instructions, tools, and state management to perform tasks autonomously.

AIAgent agent = chatClient
    .AsAIAgent(
        instructions: "You're a helpful assistant...",
        name: "MyAgent",
        tools: [...]);

Sessions

AgentSession maintains conversation history across multiple turns:

AgentSession session = await agent.CreateSessionAsync();
await agent.RunAsync("What's the weather?", session);
await agent.RunAsync("How about tomorrow?", session); // Remembers context

Tools (Function Calling)

Tools let agents call .NET functions to access external data:

AIFunctionFactory.Create(ListFlavorsTool.ListFlavors)

The agent decides when to call functions based on the conversation.

Middleware

Middleware intercepts messages before/after the agent runs:

var agentWithMiddleware = agent
    .AsBuilder()
    .Use(runFunc: FilterContentMiddleware)
    .Build();

Workflows

Orchestrate multiple agents to handle complex, multi-step tasks (see Demo 3).

Next Steps

• Add your own tools: Extend agents with custom functions
• Experiment with prompts: Modify instructions to change agent behavior
• Deploy to production: Package as a web API or Azure Function
• Explore Microsoft Foundry: Deploy and evaluate agents at scale

Resources

• Microsoft Agent Framework Documentation
• Azure OpenAI Service
• GitHub Models
• Microsoft Foundry

License

This project is licensed under the MIT License - see the LICENSE file for details.