14 KiB
A personal AI assistant system built with Java + Vue 3, powered by Spring AI Alibaba. Features multi-agent orchestration, a flexible tool/skill system with MCP protocol support, multi-layer memory, and multi-channel adapters.
Core capabilities:
Multi-Agent Orchestration — ReAct (Thought → Action → Observation loop) and Plan-and-Execute (auto-decompose complex tasks into ordered sub-steps). Create multiple independent agents, each with their own personality and tools.
Tool & Skill System — Built-in tools (web search, date/time) + MCP protocol for external tool integration. Install skill packages from ClawHub marketplace or custom sources.
Multi-Layer Memory — Short-term context window with auto-compression, event-driven post-conversation memory extraction, workspace files (PROFILE.md / MEMORY.md / daily notes), and scheduled memory consolidation.
Every Channel — Web console, DingTalk, Feishu, WeChat Work, Telegram, Discord, QQ. One MateClaw, connect as needed.
Multi-Provider Models — DashScope (Qwen), OpenAI, Ollama, DeepSeek, OpenRouter, Zhipu AI, Volcano Engine, and more. Configure in the web UI.
Desktop App — Electron-based desktop application with auto-update support. Download and double-click to run.
Table of Contents
- Quick Start
- Screenshots
- Architecture
- Tech Stack
- Features
- Documentation
- Roadmap
- Contributing
- Contact Us
- License
Quick Start
Prerequisites
- Java 17+
- Node.js 18+ & pnpm
- Maven 3.9+ (or use
mvnw) - At least one LLM API Key (e.g., DashScope)
Option 1: Local Development
1. Start the backend
cd mateclaw-server
export DASHSCOPE_API_KEY=your-key-here
mvn spring-boot:run
# Backend runs at http://localhost:18088
# H2 Console: http://localhost:18088/h2-console
# API Docs (Knife4j): http://localhost:18088/doc.html
2. Start the frontend
cd mateclaw-ui
pnpm install
pnpm dev
# Frontend runs at http://localhost:5173 (proxies /api to :18088)
3. Log in
Open http://localhost:5173 and log in with admin / admin123.
Option 2: Docker
cp .env.example .env
# Edit .env — fill in DASHSCOPE_API_KEY and other variables
docker compose up -d
# Service runs at http://localhost:18080 (MySQL + backend)
Option 3: Desktop Application
Download the installer from GitHub Releases:
| Platform | File | Notes |
|---|---|---|
| macOS (Apple Silicon) | MateClaw_<version>_arm64.dmg |
Recommended for M1/M2/M3/M4 Mac |
| macOS (Apple Silicon) | MateClaw_<version>_arm64.zip |
Alternative zip format |
| Windows (x64) | MateClaw_<version>_Setup.exe |
For most Windows PCs |
| Windows (ARM64) | MateClaw_<version>_arm64_Setup.exe |
For ARM-based Windows (e.g. Surface Pro X) |
Double-click to run. The app bundles JRE 21 + the Spring Boot backend, no Java installation needed. Supports auto-update via GitHub Releases.
macOS users: If macOS blocks the app, right-click → Open → Open again, or go to System Settings → Privacy & Security → Open Anyway.
Screenshots
Architecture
mateclaw/
├── mateclaw-server/ # Spring Boot backend
│ ├── src/main/java/vip/mate/
│ │ ├── agent/ # Agent engine (ReAct, Plan-and-Execute, StateGraph)
│ │ ├── planning/ # Task planning (Plan / SubPlan models)
│ │ ├── tool/ # Tool system (built-in + MCP adapters)
│ │ ├── skill/ # Skill management (workspace + ClawHub)
│ │ ├── channel/ # Channel adapters (Web, DingTalk, Feishu, etc.)
│ │ ├── workspace/ # Conversations, messages, workspace files
│ │ ├── memory/ # Memory extraction & consolidation
│ │ ├── llm/ # Multi-provider model configs
│ │ ├── cron/ # Scheduled tasks (CronJob)
│ │ ├── auth/ # Spring Security + JWT
│ │ └── config/ # Spring bean configurations
│ └── src/main/resources/
│ ├── application.yml # Main config (H2 for dev)
│ ├── prompts/ # Prompt templates
│ └── db/ # Schema & seed data (schema.sql, data.sql)
├── mateclaw-ui/ # Vue 3 SPA frontend
│ └── src/
│ ├── views/ # Pages (ChatConsole, AgentWorkspace, SkillMarket, etc.)
│ ├── components/ # Reusable components
│ ├── stores/ # Pinia stores (domain-driven)
│ ├── api/ # Axios HTTP client
│ ├── router/ # Vue Router
│ ├── types/ # TypeScript types
│ └── i18n/ # Internationalization (zh-CN, en-US)
├── mateclaw-desktop/ # Electron desktop app
├── docs/ # VitePress documentation (zh + en)
├── docker-compose.yml
└── .env.example
Tech Stack
| Layer | Technology |
|---|---|
| Backend Framework | Spring Boot 3.5 + Spring AI Alibaba 1.1 |
| LLM Integration | DashScope, OpenAI, Ollama, DeepSeek, OpenRouter, Zhipu, Volcano Engine |
| Agent Engine | StateGraph (ReAct + Plan-and-Execute) |
| Database | H2 (dev) / MySQL 8.0+ (prod) |
| ORM | MyBatis Plus 3.5 |
| Authentication | Spring Security + JWT |
| API Docs | Knife4j (OpenAPI 3) |
| Frontend | Vue 3 + TypeScript + Vite |
| State Management | Pinia |
| UI Components | Element Plus |
| Styling | TailwindCSS 4 |
| Desktop | Electron + electron-updater |
| Docs Site | VitePress |
Features
Agent System
- ReAct Agent — Thought → Action → Observation reasoning loop with tool calling
- Plan-and-Execute — Auto-decompose complex tasks into ordered sub-steps with progress tracking
- Dynamic Agent — Load agent configs from database at runtime
- Multi-Agent — Create multiple independent agents, each with their own system prompt, tools, and personality
Tool & Skill System
- Built-in Tools — Web search (Serper/Tavily), date/time, workspace memory read/write
- MCP Protocol — Connect external tools via Model Context Protocol (stdio and SSE transports)
- Skill Packages — Install/uninstall skill packages with
SKILL.mdmanifests - ClawHub Marketplace — Browse and install skills from the ClawHub registry
- Workspace Skills — Convention-based skill directory at
~/.mateclaw/skills/{name}/
Memory System
- Short-Term — Conversation context window with auto-compression when token budget exceeded
- Post-Conversation Extraction — Event-driven async LLM analysis, writes to PROFILE.md / MEMORY.md / daily notes
- Memory Consolidation — Scheduled daily emergence (CronJob at 2:00 AM) merges daily notes into long-term memory
- Workspace Files — Per-agent AGENTS.md, SOUL.md, PROFILE.md, MEMORY.md, memory/*.md
- Agent Memory Tool — Agents can read/write their own workspace files during conversations
Multi-Channel
- Web Console — SSE streaming with rich message rendering (Markdown, code, plans)
- DingTalk — Webhook + event subscription
- Feishu (Lark) — Webhook + event subscription
- WeChat Work — Callback API
- Telegram — Bot API with webhook
- Discord — Bot with slash commands
- QQ — QQ Bot API
Model Providers
Configure in the web UI (Settings → Models). Supported providers:
| Provider | Models |
|---|---|
| DashScope (Alibaba) | Qwen-Max, Qwen-Plus, Qwen-Turbo, Qwen-Long, QVQ |
| OpenAI | GPT-4o, GPT-4o-mini, o1, o3 |
| DeepSeek | DeepSeek-Chat, DeepSeek-Reasoner |
| Ollama | Any locally-served model |
| OpenRouter | Access 200+ models via unified API |
| Zhipu AI | GLM-4-Plus, GLM-4-Flash |
| Volcano Engine | Doubao-Pro, Doubao-Lite |
| SiliconFlow | DeepSeek, Qwen via SiliconFlow |
Security
- Spring Security + JWT — Token-based authentication
- Tool Guard — Approval rules for sensitive tool operations
- File Validation — Path traversal prevention for workspace files
- Skill Security — Validation during skill installation
Scheduled Tasks
- CronJob System — Create scheduled tasks with 5-field cron expressions
- Memory Consolidation — Auto-triggered daily for each agent
- Custom Tasks — Schedule any prompt to run periodically
Documentation
| Topic | Description |
|---|---|
| Introduction | What MateClaw is and core concepts |
| Quick Start | Install and run (local, Docker, desktop) |
| Console | Web UI: chat and agent configuration |
| Agents | Agent engine: ReAct, Plan-and-Execute, StateGraph |
| Models | Configure cloud, local, and custom providers |
| Tools | Built-in tools and custom tool development |
| Skills | Skill packages and ClawHub marketplace |
| MCP | Model Context Protocol integration |
| Memory | Multi-layer memory system |
| Channels | DingTalk, Feishu, Telegram, Discord, and more |
| Security | Authentication and tool guard |
| Desktop | Desktop application guide |
| API Reference | REST API documentation |
| Configuration | Configuration reference |
| FAQ | Common questions and troubleshooting |
Roadmap
| Area | Item | Status |
|---|---|---|
| Agent | Multi-agent collaboration and delegation | Planned |
| Agent | Multimodal input (image, audio, video) | Planned |
| Models | Small + large model routing | Planned |
| Memory | Vector DB long-term memory (RAG) | Planned |
| Memory | Multimodal memory fusion | Planned |
| Skills | Richer ClawHub ecosystem | In Progress |
| Channels | WeChat personal (iLink Bot) | Planned |
| Channels | Email channel | Planned |
| Desktop | Linux support | Planned |
| Security | Multi-tenant support | Planned |
| Console | Plugin marketplace in web UI | Planned |
Status: In Progress — actively being worked on; Planned — queued or under design.
Contributing
MateClaw is open to contributions! Whether it's bug fixes, new features, documentation improvements, or new channel/tool integrations — all contributions are welcome.
# Clone the repository
git clone https://github.com/matevip/mateclaw.git
cd mateclaw
# Backend
cd mateclaw-server
mvn clean compile
# Frontend
cd ../mateclaw-ui
pnpm install
pnpm dev
Please read CONTRIBUTING.md (if available) before submitting a PR.
Contact Us
| Discord | X (Twitter) | DingTalk |
|---|---|---|
| Coming soon | Coming soon | Coming soon |
Why MateClaw?
Mate — a companion, always by your side. Claw — sharp, capable, ready to grab any task. MateClaw is your personal AI mate that lends a claw whenever you need it. Built as a monolith with modular design, it's easy to deploy, extend, and customize.
License
MateClaw is released under the Apache License 2.0.