Production-Ready AI Agent Development Framework Powered by Claude Agent SDK
基于 Claude Agent SDK 构建的生产级 AI 智能体开发框架

English | 中文

Agent Kit is a comprehensive AI agent development framework that integrates Claude Agent SDK, providing a complete solution from frontend to backend. This project aims to help developers quickly build, deploy, and scale production-grade AI Agent applications. It includes built-in multi-channel access via WebSocket, Discord, and Telegram, with unified session routing and message handling.

• Introduction
• Architecture
• Quick Start
• Project Structure
• Core Features
• Configuration
• Third-Party IM Integration
• API Documentation
• Development Guide
• Contributing
• License

• Python: 3.11 or higher
• Node.js: 24.0 or higher
• Docker & Docker Compose: Latest version
• Agent API Key: Get from Anthropic or Bigmodel

1️⃣ Clone the repository

git clone https://github.com/leemysw/agent-kit.git
cd agent-kit

2️⃣ Configure environment variables

# Copy environment variable template
cp example.env .env
# Edit .env file and add your API key

3️⃣ Start services

make start
╰─ make start
TAG=0.1.2 docker compose -f deploy/docker-compose.yml up -d
[+] Running 3/3
 ✔ Container deploy-agent-kit-1  Started                                                                                                                                           1.8s 
 ✔ Container deploy-web-1   Started                                                                                                                                           0.9s 
 ✔ Container deploy-nginx-1      Running                                                                                                                                           0.0s 

✅ Agent Kit is running!
🌐 Web UI: http://localhost
📋 Logs: run 'make logs' to view service logs

4️⃣ Access the application

• Application URL: http://localhost

1️⃣ Clone the repository

git clone https://github.com/leemysw/agent-kit.git
cd agent-kit

2️⃣ Backend setup

# Install Python dependencies
pip install -r agent/requirements.txt

# Configure environment variables
cp example.env .env
# Edit .env file and add your API key

Configure .env file:

# Claude API configuration
ANTHROPIC_API_KEY=your_api_key_here
ANTHROPIC_BASE_URL=https://api.anthropic.com or https://open.bigmodel.cn/api/anthropic
ANTHROPIC_MODEL=claude-3-5-sonnet-20241022 or glm-5

# Server configuration
HOST=0.0.0.0
PORT=8010
DEBUG=true
WORKERS=1

Initialize database:

# Run database migrations to create tables
alembic upgrade head

3️⃣ Frontend setup

cd web

# Install dependencies
npm install

# Configure environment variables
cp example.env .env.local
# Edit .env.local file

Configure .env.local file:

# Development environment configuration
NEXT_PUBLIC_API_URL=http://localhost:8010/agent/v1
NEXT_PUBLIC_WS_URL=ws://localhost:8010/agent/v1/chat/ws
NEXT_PUBLIC_DEFAULT_CWD=/opt/app/playground
NEXT_PUBLIC_DEFAULT_MODEL=glm-5

4️⃣ Run the project

# Start backend (in project root directory)
python main.py

# Start frontend (in web directory)
npm run dev

5️⃣ Access the application

• Application URL: http://localhost:3000

agent-kit/
├── agent/                         # Backend service
│   ├── api/                       # API routes
│   ├── core/                      # Core configuration
│   ├── service/                   # Business logic
│   │   ├── websocket_handler.py   # WebSocket handler
│   │   └── session_manager.py     # Session management
│   ├── shared/                    # Shared modules
│   └── utils/                     # Utility functions
├── web/                           # Frontend application
│   ├── src/
│   │   ├── app/                   # Next.js pages
│   │   ├── components/            # React components
│   │   ├── hooks/                 # Custom Hooks
│   │   ├── lib/                   # Utility library
│   │   ├── store/                 # Zustand state management
│   │   └── types/                 # TypeScript types
├── alembic/                       # Database migrations
├── deploy/                        # Deployment files
├── docs/                          # Documentation
│   ├── websocket-session-flow.md  # WebSocket flow
│   └── guides/                    # Claude Agent SDK guides
├── main.py                        # Application entry point
└── README.md                      # This file

• ✅ WebSocket real-time communication
• ✅ Streaming response support
• ✅ Session persistence
• ✅ Message history management

• ✅ Multi-session support
• ✅ Session search and filtering

• ✅ Claude Agent SDK integration
• ❌ Custom tool calling (in development)
• ❌ Slash command system (in development)
• ❌ Skills system (in development)
• ❌ MCP protocol support (in development)

• ✅ Fine-grained tool permission control
• ✅ User confirmation mechanism

The backend currently supports three message entry channels:

• WebSocket (Web UI)
• Discord (agent/service/channel/discord_channel.py)
• Telegram (agent/service/channel/telegram_channel.py)

# Discord
DISCORD_ENABLED=true
DISCORD_BOT_TOKEN=your_discord_bot_token
DISCORD_ALLOWED_GUILDS=123456789012345678,987654321098765432
DISCORD_TRIGGER_WORD=@agent-kit

# Telegram
TELEGRAM_ENABLED=true
TELEGRAM_BOT_TOKEN=your_telegram_bot_token
TELEGRAM_ALLOWED_USERS=12345678,87654321

Channels are registered automatically at app startup based on DISCORD_ENABLED/TELEGRAM_ENABLED.

All IM channels share the same session key pattern:

agent:<agentId>:<channel>:<chatType>:<ref>[:topic:<threadId>]

Examples:

• Discord group chat: agent:main:dg:group:<guild_id>:<channel_id>
• Telegram direct message: agent:main:tg:dm:<user_id>

• Discord not responding: ensure Message Content Intent is enabled for the bot.
• Telegram not receiving messages: check privacy mode and make sure the current user is included in TELEGRAM_ALLOWED_USERS.
• DISCORD_TRIGGER_WORD: the current implementation strips the trigger word if present, but does not require it.

For detailed guides and API documentation, please visit:

• Frontend API Documentation - React components, types, and API interfaces
• WebSocket Session Flow - WebSocket session and data flow
• Guides - Comprehensive guides for various features

• Session Management - Session creation, management, and message handling
• Streaming vs Single Mode - AI response mode comparison
• Custom Tools - Creating and using custom AI tools
• Slash Commands - Custom slash command development
• Skills Guide - Skill system usage and development
• MCP Integration - Model Context Protocol integration
• Hosting Guide - Production deployment and configuration
• Permissions Management - Permission control and security settings

All forms of contributions are welcome!

If you find a bug or have a new feature suggestion, please submit it through GitHub Issues.

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

• Claude Agent SDK - Core AI framework