Microsoft Agent Framework Samples

A Blazor Web Application built on .NET 10 that showcases 24 interactive samples demonstrating the Microsoft Agent Framework (Microsoft.Extensions.AI.Agents) across 6 categories — from basic request/response through advanced prompting strategies.

Live Demo: Run the app and navigate to http://localhost:5xxx to explore the interactive samples.

---

✨ Features

• 24 interactive agent samples organized in 6 categories
• Modern, responsive UI built with Bootstrap 5 and a central CSS design system
• Real-time streaming responses
• Multi-turn conversation threads
• Function/tool calling with and without user approval
• Structured JSON output extraction
• Persisted conversation storage
• Multimodal image analysis
• Remote MCP (Model Context Protocol) server integration
• Background/async response processing
• Middleware pipeline (logging, function override, content redaction)
• Declarative agent configuration via YAML
• Advanced reasoning with OpenAI o4-mini
• 7 prompting strategy demonstrations (CoT, ToT, GoT, PoT, ReAct, Reflection, Self-Refine)

---

🛠 Technology Stack

Technology	Details
.NET 10 / Blazor Web App	Server-Side Rendering + Interactive Server components
Microsoft Agent Framework	Microsoft.Extensions.AI.Agents
Bootstrap 5	Responsive UI framework
OpenAI API (via OpenRouter)	OpenAI-compatible model access
Semantic Kernel	Plugin and middleware integration samples

---

📋 Prerequisites

• .NET 10 SDK
• An API key from OpenRouter (supports OpenAI-compatible models)

---

⚙️ Configuration

Set your OpenRouter API key using one of the following methods:

Option 1 — appsettings.json

Edit 0-Agents/AgentsWebUI/appsettings.json:

{
  "OpenRouter": {
    "ApiKey": "your-api-key-here"
  }
}

Option 2 — Environment variable

OPEN_ROUTER_API_KEY=your-api-key-here

---

🚀 Running the Application

cd 0-Agents/AgentsWebUI
dotnet run

Then open your browser to the URL shown in the terminal output (e.g., http://localhost:5xxx).

---

📚 Sample Categories

🚀 Basics (3 samples)

Agent	Name	Description
Agent 01	Basic Agent	Demonstrates creating and running an agent. Shows simple request/response and streaming response.
Agent 02	Agent with Thread	Multi-turn conversations using threads. The agent remembers context of previous messages.
Agent 05	Structured Output	Extracts structured JSON data from unstructured text. Populates a PersonInfo object from natural language.

---

🔧 Functions & Tools (5 samples)

Agent	Name	Description
Agent 03	Agent with Function/Tool	Provides Add and Multiply functions to an agent to solve math problems.
Agent 04	Functions with User Approval	Adds a user-approval step before any function is executed.
Agent 10	Agent as Function	Uses one agent as a callable tool for another agent.
Agent 13	Plugins	Plugin integration adding external services (weather, current time) via dependency injection.
Agent 24	Skills	Provides a UI for configuring and running an agent with file-based skills via FileAgentSkillsProvider, demonstrating custom instructions/message panels and streaming responses.

---

💾 Storage & Persistence (3 samples)

Agent	Name	Description
Agent 06	Persisted Conversation	Serializes a conversation thread to disk and resumes it in a later interaction.
Agent 07	Custom Thread Storage	Uses a custom VectorChatMessageStore to persist conversation history in memory.
Agent 11	Chat Reduction	Chat reduction techniques using message store reducers to limit context size.

---

🌐 Multimodal & External Services (3 samples)

Agent	Name	Description
Agent 08	Agent Using Image	Multimodal model analyzes an image from a URL using a text prompt.
Agent 09	Using Remote MCP Server	Connects to the Microsoft Learn MCP server to answer questions about Microsoft technologies.
Agent 12	Background Responses	Asynchronous background responses with response resumption and continuous streaming.

---

⚡ Advanced Features (3 samples)

Agent	Name	Description
Agent 14	Middleware	Middleware for logging, function overriding, and content redaction.
Agent 15	Declarative Agent	Agent configured via a YAML definition with an output schema returning structured JSON.
Agent 16	Reasoning	Advanced reasoning with OpenAI's o4-mini model, exposing step-by-step reasoning content.

---

🧠 Prompting Strategies (7 samples)

Agent	Name	Description
Agent 17	Chain-of-Thought (CoT)	Step-by-step reasoning that guides the model to think through problems logically before answering.
Agent 18	Tree of Thoughts (ToT)	Explores multiple reasoning branches in parallel and selects the most promising solution.
Agent 19	Graph of Thoughts (GoT)	Builds a flexible reasoning graph where thoughts can merge, branch, and reference each other.
Agent 20	Program of Thoughts (PoT)	Generates algorithmic pseudocode to separate computation from reasoning.
Agent 21	ReAct	Interleaves reasoning traces with tool actions in a Thought→Action→Observation loop.
Agent 22	Reflection	Generates an initial response then applies self-criticism to identify improvements.
Agent 23	Self-Refine	Iteratively improves output through multiple critique-and-refine rounds.

---

🧠 Workflows (3 samples)

Workflow	Name	Description
Workflow 01	Translation Chain	Translates a user’s message sequentially from French → Spanish → English using a chain of translation agents.
Workflow 02	Group Chat Tool Approval	Runs a group chat workflow between QA and DevOps agents that requires user approval for deployment tools.
Workflow 03	Workflow As Agent	Executes a workflow as a single agent by concurrently running French and English language agents and aggregating their responses.

---

📁 Project Structure

agent-framework-sample/
├── 0-Agents/
│   └── AgentsWebUI/
│       ├── Components/
│       │   ├── Layout/          # MainLayout, NavMenu
│       │   └── Pages/           # Home
│       │       ├── Agents/       # Agent01-Agent24
│       │       └── Workflows/    # Workflow01-Workflow03
│       ├── wwwroot/
│       │   └── app.css          # Central design system CSS
│       ├── Program.cs
│       └── appsettings.json
├── plans/                       # UI design documentation
└── readme.md

---

📄 License

This project is licensed under the MIT License. See LICENSE for details.