Multi-Agent Architecture Platform

🌍 Language Select | 语言选择

🚀 An enterprise-grade multi-agent architecture platform based on Spring AI and WebFlux, supporting advanced paradigms like Function Calling and ReAct, providing powerful and flexible infrastructure for building intelligent conversational systems.

✨ Core Features

🤖 Multi-Agent Collaboration Architecture - Unified management and scheduling of various types of intelligent agents
🔄 Function Calling Support - Complete tool calling capabilities with custom function extensions
⚡ ReAct Paradigm Implementation - Reasoning-Action cycles enabling complex multi-step reasoning
🌊 Reactive Programming - Non-blocking asynchronous architecture based on WebFlux
💾 Multi-Data Source Support - Dual data storage strategy with PostgreSQL + Redis
🐳 Containerized Deployment - Complete Docker support with one-click deployment
🔧 Highly Configurable - Flexible configuration management supporting multiple LLM providers
🎯 Enterprise Features - User isolation, permission management, session persistence

🏗️ Architecture Overview

graph TB
    A[User Request] --> B[Assistant Controller]
    B --> C[Intent Agent]
    C --> D[Agent Manager]
    D --> E[Specific Agent Instance]
    E --> F[LLM Service]
    F --> G[Function Calling]
    G --> H[Tool Execution]
    G --> I[Card Rendering]
    H --> F
    I --> J[Response Return]

    subgraph "Data Layer"
        K[PostgreSQL]
        L[Redis]
    end

    D -.-> K
    D -.-> L

🚀 Quick Start

Requirements

Java: 17+
Maven: 3.6+
PostgreSQL: 12+
Redis: 6+

Local Development Setup

Clone Project

git clone https://github.com/your-username/multi-agent-arch.git
cd multi-agent-arch

Configure Database

Create PostgreSQL database:
```
CREATE DATABASE assistant;
```

Configure Environment Variables

Copy and edit configuration file:

# Database configuration
export SPRING_DATASOURCE_URL=jdbc:postgresql://localhost:5432/assistant
export SPRING_DATASOURCE_USERNAME=your_username
export SPRING_DATASOURCE_PASSWORD=your_password

# Redis configuration
export SPRING_REDIS_HOST=localhost
export SPRING_REDIS_PORT=6379
export SPRING_REDIS_PASSWORD=your_redis_password

# LLM configuration (OpenAI compatible API)
export LLM_MODEL_SERIES=qwen
export LLM_API_KEY=your_api_key
export LLM_ADDRESS=https://dashscope.aliyuncs.com/compatible-mode/v1
export LLM_MODEL=qwen-plus

Run Application

mvn spring-boot:run

Or build and run:

mvn clean package
java -jar target/mss-ai.jar

Verify Service

After service startup, visit health check endpoint:
```
curl http://localhost:8080/ai/actuator/health
```

Docker Deployment

Build Image

mvn clean package
docker build -t multi-agent-arch .

Run Container

docker run -d \
  --name multi-agent-arch \
  -p 8080:8080 \
  -e SPRING_DATASOURCE_URL=jdbc:postgresql://host.docker.internal:5432/assistant \
  -e SPRING_DATASOURCE_USERNAME=your_username \
  -e SPRING_DATASOURCE_PASSWORD=your_password \
  -e LLM_API_KEY=your_api_key \
  multi-agent-arch

📚 API Documentation

Chat Completion Interface

POST /ai/chat/completions

Initiate conversation with agents, supporting streaming responses.

curl -X POST http://localhost:8080/ai/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Help me check today'\''s weather",
    "extra": {
      "location": "Beijing",
      "language": "zh-CN"
    }
  }'

Request Parameters:

content (string, required): User input message content
extra (object, optional): Additional context information

Response Format:

{
  "content": "OK, let me help you check the weather information in Beijing...",
  "status": "REPLYING",
  "data": {},
  "design": {}
}

Intent Recognition

The system automatically recognizes user intents and dispatches corresponding agents:

General Conversation Agent - Handles daily conversations
Tool Calling Agent - Executes specific function calls
Knowledge Retrieval Agent - Performs information queries and analysis

Function Calling Support

The system supports complete function calling lifecycle:

Function Registration - Define available tool functions in agent configuration
Call Parsing - LLM automatically identifies functions to call
Parameter Extraction - Extract function call parameters from conversations
Function Execution - Call actual business logic
Result Processing - Feed execution results back to LLM for next step processing

🛠️ Core Components

Agent Manager

Responsible for agent lifecycle management:

Agent type loading and management
Agent instance creation and destruction
Session state timeout management (15-minute auto-expiration)

Intent Agent

LLM-based intent recognition system:

Analyze business type of user input
Auto-dispatch to corresponding professional agents
Support dynamic intent configuration

LLM Service

Unified LLM calling interface:

Support multiple LLM providers (currently implements Qwen)
Streaming and non-streaming response modes
Complete Function Calling support

Function Call Service

Extensible function calling framework:

Support local and remote function calls
Unified calling interface and response format
Support card-style response rendering

🔧 Configuration

Complete application.yaml Example

# Basic service configuration
spring:
  webflux:
    base-path: /ai
  data:
    redis:
      host: ${SPRING_REDIS_HOST:localhost}
      port: ${SPRING_REDIS_PORT:6379}
      password: ${SPRING_REDIS_PASSWORD}
  datasource:
    url: ${SPRING_DATASOURCE_URL:jdbc:postgresql://localhost:5432/assistant}
    username: ${SPRING_DATASOURCE_USERNAME}
    password: ${SPRING_DATASOURCE_PASSWORD}
    driver-class-name: org.postgresql.Driver
  jpa:
    hibernate:
      ddl-auto: update
    show-sql: true
    properties:
      hibernate:
        dialect: org.hibernate.dialect.PostgreSQLDialect

# Business configuration
business:
  llm:
    model-series: ${LLM_MODEL_SERIES:qwen}
    api-key: ${LLM_API_KEY}
    url: ${LLM_ADDRESS}
    model: ${LLM_MODEL}

# Logging configuration
logging:
  level:
    root: ${LOGGING_LEVEL:info}
    site.nullpointer.mss.ai: DEBUG

🎯 Usage Examples

1. Basic Conversation

curl -X POST http://localhost:8080/ai/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Hello, please introduce yourself"
  }'

2. Function Calling Example

curl -X POST http://localhost:8080/ai/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Help me create a todo: complete project documentation",
    "extra": {
      "priority": "high",
      "category": "work"
    }
  }'

3. ReAct Paradigm Example

curl -X POST http://localhost:8080/ai/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "content": "I need to analyze user purchase behavior data, can you help me?"
  }'

The system will automatically execute the following steps:

Analyze user requirements
Call data analysis tools
Parse analysis results
Provide insights and recommendations

🧪 Development Guide

Adding New Agent Types

Insert new agent type in database:

INSERT INTO agent_type (type, system_prompt, function_tools, temperature, description, is_public, enterprise_id)
VALUES (
  'CUSTOM_AGENT',
  'You are a professional data analysis assistant...',
  '[{"function": {"name": "analyze_data", "description": "Analyze data"}}]',
  0.7,
  'Data Analysis Agent',
  true,
  'default'
);

Restart application or call AgentManager's loadAgentType() method to reload configuration.

Custom Function Calling

Implement FunctionCallService interface:

@Service
public class CustomFunctionCallService implements FunctionCallService {
    @Override
    public FunctionCallResult apply(FunctionCallRequest request) {
        // Implement your business logic
        return FunctionCallResult.builder()
            .content("Execution result")
            .build();
    }
}

{
  "function": {
    "name": "custom_function",
    "description": "Custom function description",
    "parameters": {
      "type": "object",
      "properties": {
        "param1": {"type": "string"}
      }
    }
  }
}

🤝 Contributing

We welcome all forms of contributions! Please read the following guidelines:

Fork this repository
Create your feature branch (git checkout -b feature/AmazingFeature)
Commit your changes (git commit -m 'Add some AmazingFeature')
Push to the branch (git push origin feature/AmazingFeature)
Open a Pull Request

Code Standards

Follow Spring Boot best practices
Use Lombok to reduce boilerplate code
Keep code clean and readable
Add appropriate unit tests

📄 License

This project is licensed under the Apache License 2.0. See LICENSE file for details.

🙏 Acknowledgments

Spring AI - Providing powerful AI integration framework
Spring WebFlux - Reactive web framework
Qwen - Providing excellent LLM capabilities

📞 Contact

Project homepage: https://github.com/your-username/multi-agent-arch
Issue feedback: Issues
Discussions: Discussions

If this project helps you, please give us a ⭐️

Made with ❤️ by the Multi-Agent Architecture Team

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Multi-Agent Architecture Platform

🌍 Language Select | 语言选择

✨ Core Features

🏗️ Architecture Overview

🚀 Quick Start

Requirements

Local Development Setup

Docker Deployment

📚 API Documentation

Chat Completion Interface

Intent Recognition

Function Calling Support

🛠️ Core Components

Agent Manager

Intent Agent

LLM Service

Function Call Service

🔧 Configuration

Complete application.yaml Example

🎯 Usage Examples

1. Basic Conversation

2. Function Calling Example

3. ReAct Paradigm Example

🧪 Development Guide

Adding New Agent Types

Custom Function Calling

🤝 Contributing

Code Standards

📄 License

🙏 Acknowledgments

📞 Contact

FilesExpand file tree

README_EN.md

Latest commit

History

README_EN.md

File metadata and controls

Multi-Agent Architecture Platform

🌍 Language Select | 语言选择

✨ Core Features

🏗️ Architecture Overview

🚀 Quick Start

Requirements

Local Development Setup

Docker Deployment

📚 API Documentation

Chat Completion Interface

Intent Recognition

Function Calling Support

🛠️ Core Components

Agent Manager

Intent Agent

LLM Service

Function Call Service

🔧 Configuration

Complete application.yaml Example

🎯 Usage Examples

1. Basic Conversation

2. Function Calling Example

3. ReAct Paradigm Example

🧪 Development Guide

Adding New Agent Types

Custom Function Calling

🤝 Contributing

Code Standards

📄 License

🙏 Acknowledgments

📞 Contact