# Admin.Net.AI **Repository Path**: micropu/admin.net.ai ## Basic Information - **Project Name**: Admin.Net.AI - **Description**: Admin.NET.Ai 是深度集成 MEAI 与 MAF 框架一个目前集合AI所有能力的合集，包括15+场景模板项能力，全部在控制台项目中 - **Primary Language**: C# - **License**: Not specified - **Default Branch**: master - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 9 - **Created**: 2026-03-23 - **Last Updated**: 2026-03-23 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README # Admin.NET.Ai
**English** | **[中文](./README.md)** [![.NET](https://img.shields.io/badge/.NET-10.0-512BD4?logo=dotnet)](https://dotnet.microsoft.com/) [![License](https://img.shields.io/badge/License-MIT-green.svg)](LICENSE) [![MEAI](https://img.shields.io/badge/Microsoft.Extensions.AI-✓-blue)](https://www.nuget.org/packages/Microsoft.Extensions.AI)
--- Admin.NET.Ai is an enterprise-grade AI capability core library built on **.NET 10**. It uses a pipeline/middleware architecture and deeply integrates `Microsoft.Extensions.AI` (MEAI) and `Microsoft.Agents` frameworks. > [!IMPORTANT] > **Requirements**: This project requires .NET 10 features. Recommend using [mise](https://mise.jdx.dev/) to manage the environment with `mise use dotnet` to activate the SDK. --- ## ✨ Core Features | Feature | Description | | :--- | :--- | | 🔌 **Multi-Provider Support** | Seamlessly switch between OpenAI, DeepSeek, Qwen, Gemini, Ollama, etc. | | 🤖 **Multi-Agent Orchestration** | Sequential/Parallel/Orchestrator/Roundtable modes, LLM-driven Agent auto-discovery & dispatch | | 🔧 **Enhanced Tool System** | FileSystem/Search/Shell tools + self-managed approval + MCP tool discovery | | 🎨 **Media Generation** | TTS/ASR/Image/Video generation with multi-provider support | | 📚 **Hybrid RAG** | Vector retrieval + Neo4j GraphRAG + Reranking | | ⚡ **Three-Layer Middleware** | Chat pipeline (Token/cost/caching) + Tool pipeline (approval/monitoring/validation) + rate limiting/audit | | 🗜️ **Three-Zone Compression** | First-turn preservation + LLM summarization of history + recent message retention | | 🔥 **Hot-Reload Scripting** | Natasha C# script engine for dynamic Agent logic updates | | 📊 **Full Observability** | Trace timeline + DevUI visual debugging | ### 📋 Feature Demos Overview (Console) | # | Category | Module | Description | | :---: | :--- | :--- | :--- | | **★1** | **Universal** | **Universal Agent** | **All-in-One Agent, auto-loads all tools & agents** | | 2 | Chat Basics | Basic Chat & Middleware | Chat, Audit, Tokens | | 3 | Chat Basics | Prompt Engineering | Prompt Templates | | 4 | Chat Basics | Structured Extraction | JSON Schema, TOON | | 5 | Chat Basics | Code Generator | Structured Output | | 6 | Chat Basics | Multimodal | Vision & Audio | | 7 | Tools | Smart Tools & Approval | Discover, Approval | | 8 | Tools | Enhanced Tool System | FileSystem/Search/Shell | | 9 | Tools | MCP Protocol | External Tool Integration | | 10 | Tools | MCP Calendar Assistant | Official SDK Tool Calls | | 11 | Tools | MCP MiniApi Server | External Tool Integration | | 12 | Agents | Built-in Agents | Sentiment/Knowledge Graph/Quality | | 13 | Agents | LLM Agent Dispatch | Auto-Discovery | | 14 | Agents | Multi-Agent Workflow | MAF Sequential & Autonomous | | 15 | Agents | Multi-Agent Review | Writer→Reviewer→Editor | | 16 | Agents | Customer Service Routing | Intent Recognition + Routing | | 17 | Data | RAG Knowledge Retrieval | GraphRAG & Vector | | 18 | Data | RAG + Agent QA | Knowledge Base + Reasoning | | 19 | Data | Context Compression | ThreeZone/Summarizing/Counting | | 20 | Data | Conversation Persistence | Thread & Database | | 21 | Infra | Middleware Deep-dive | Middleware Stack | | 22 | Infra | Content Safety | Sensitive Word Replace + PII Masking | | 23 | Infra | Monitoring & Metrics | OpenTelemetry | | 24 | Infra | Storage Strategies | Hot/Cold/Vector | | 25 | Infra | Dynamic Script Hot-Reload | Natasha Scripting | | 26 | Scenarios | Real-world Scenarios | Comprehensive Applications | | 27 | Scenarios | Media Generation | TTS/ASR/Image/Video | --- ## 🚀 Quick Start ### 1. Install ```bash dotnet add package Admin.NET.Ai # Not uploaded yet, add project reference manually ``` ### 2. Register Services ```csharp services.AddAdminNetAi(configuration); ``` ### 3. Usage Examples #### Basic Chat ```csharp var aiFactory = sp.GetRequiredService(); var client = aiFactory.GetDefaultChatClient(); var response = await client.GetResponseAsync("Hello, I'm Admin.NET"); ``` #### Enhanced Tool System (Auto-Discovery + Context Injection) ```csharp var toolManager = sp.GetRequiredService(); var context = new ToolExecutionContext { WorkingDirectory = Directory.GetCurrentDirectory(), UserId = "user-001" }; var functions = toolManager.GetAllAiFunctions(context); // Auto-scan all tools with context ``` #### Multi-Agent Collaboration ```csharp var orchestrator = new EnhancedMultiAgentOrchestrator(aiFactory); orchestrator .AddAgent("Tech Expert", "Analyze from technical perspective", provider: "qwen") .AddAgent("Product Manager", "Analyze from product perspective", provider: "deepseek"); await foreach (var evt in orchestrator.RunDiscussionAsync("Impact of AI on development", rounds: 2)) { Console.Write(evt.Content); } ``` #### MCP Tools ```csharp [McpTool("Get weather information")] public WeatherInfo GetWeather([McpParameter("City name")] string city) { return new WeatherInfo { City = city, Temperature = 20 }; } ``` --- ## 🏗️ Architecture ``` Admin.NET.Ai/ ├── Abstractions/ # Interfaces: IAiFactory, IAiAgent, IAiCallableFunction, IChatReducer ├── Core/ # AiFactory, PipelineBuilder ├── Middleware/ │ ├── TokenMonitoringMiddleware # Chat pipeline: Token/cost/newline fix │ ├── ToolValidationMiddleware # Tool pipeline: permission/approval/params/sandbox/sanitize │ └── ToolMonitoringMiddleware # Tool pipeline: classification logging/duration ├── Services/ │ ├── Tools/ # ToolManager + FileSystem/Search/Shell/AgentDispatch │ ├── Context/ # ChatReducerFactory (ThreeZone/Summarizing/MessageCounting) │ ├── MCP/ # MCP Protocol + Tool Discovery │ ├── Media/ # TTS/ASR/ImageGen/VideoGen │ ├── Rag/ # Vector + GraphRAG │ └── Workflow/ # Multi-Agent Orchestration Engine ├── Configuration/ # JSON config files ├── _doc/ # User documentation └── _doc_Pro/ # Technical deep-dive ``` ### Middleware Responsibility Split ``` Request → TokenMonitoringMiddleware (Token/Cost) ↓ LLM decides to call tool ↓ ToolValidationMiddleware (Permission → Approval → Params → Sandbox → Sanitize) ↓ ToolMonitoringMiddleware (🔧Tool / 🤖Agent / ⚡Skill classification) ↓ Actual Tool Execution ``` --- ## ⚙️ Configuration ### LLMAgent.Clients.json (Providers) ```json { "LLM-Clients": { "DefaultProvider": "qwen-plus", "Clients": { "qwen-plus": { "Provider": "Qwen", "ModelId": "qwen-plus", "ApiKey": "sk-xxx" }, "deepseek": { "Provider": "DeepSeek", "ModelId": "deepseek-chat", "ApiKey": "sk-xxx" } } } } ``` ### LLMAgent.Mcp.json (MCP Servers) ```json { "LLM-Mcp": { "Servers": [ { "Name": "Filesystem", "Url": "http://localhost:3001/sse" } ] } } ``` --- ## 📖 Documentation - **User Docs**: `_doc/` - Feature introductions and usage examples - **Technical Deep-Dive**: `_doc_Pro/` - Implementation details and source code analysis --- ## 🎯 Demos Run the console demo: ```bash dotnet run --project HeMaCupAICheck ``` Select **1** to enter the Universal Agent — all tools and agents auto-loaded, LLM decides autonomously what to invoke. 27 feature demos total, grouped by category: Chat Basics · Tool System · Agent/Workflow · Data & Knowledge · Infrastructure · Scenarios. --- ## ⚖️ License Admin.NET.Ai is released under the [MIT License](LICENSE).