MateClaw

MateClaw is a personal AI operating system built with Java + Vue 3 and powered by Spring AI Alibaba.

It is not just a chat box. It is a system for building AI workers that can reason, use tools, remember context, search the live web, digest knowledge into structured Wiki pages, generate media, and show up across the channels where work actually happens.

The idea is simple:

• Give each agent a clear role
• Give it the right tools and guardrails
• Let it keep memory instead of starting from zero
• Let it work across chat, channels, documents, and media
• Keep the whole system deployable by one team without turning into infrastructure theater

MateClaw sits at the intersection of four product ideas:

1. An AI console for direct interaction
2. An agent runtime for structured reasoning and tool use
3. A knowledge system that turns raw information into reusable memory and Wiki pages
4. A deployment surface that spans web, desktop, and external channels

Most products do one of these. MateClaw is designed to make them work as one system.

MateClaw supports ReAct and Plan-and-Execute agents so the model can break work down, call tools, observe results, and continue instead of stopping at a polished paragraph.

Raw files are useful, but structured knowledge is better. MateClaw includes an LLM Wiki knowledge base that converts source materials into linked Wiki pages with summaries, backlinks, and on-demand retrieval.

Conversations should not disappear. MateClaw combines short-term context management, post-conversation extraction, workspace memory files, and scheduled consolidation so agents can build continuity over time.

Powerful tools without boundaries are a liability. MateClaw includes tool guard rules, approval flows, file-path protection, and runtime filtering so capability does not become recklessness.

A useful assistant cannot be trapped in one web page. MateClaw connects to desktop, browser, and external messaging/work channels so the agent can meet users where decisions are being made.

• A persistent assistant with memory, tools, and workspace files
• A desktop app with bundled backend and auto-update
• A web console for direct chat, planning, and configuration

• Ingest notes, documents, PDFs, and DOCX files
• Turn source materials into structured Wiki pages
• Let agents search, summarize, and read knowledge on demand

• Agents that search the web, read files, use MCP tools, and execute workflows
• Role-specific skill packages with SKILL.md
• Approval and security controls for sensitive actions

• Text-to-speech
• Speech-to-text
• Music generation
• Image generation
• Video generation

• Web console
• DingTalk
• Feishu
• WeChat Work
• Telegram
• Discord
• QQ

• ReAct agents for thought → action → observation loops
• Plan-and-Execute agents for decomposing complex work into ordered steps
• Dynamic agent configuration loaded at runtime
• Multi-agent setup with separate prompts, personalities, and tool scopes
• Runtime resilience including context pruning, smart truncation, stale stream cleanup, and recovery for longer tasks

• LLM Wiki knowledge base for structured, linked, AI-digested knowledge
• Workspace memory files such as AGENTS.md, SOUL.md, PROFILE.md, MEMORY.md, and daily notes
• Post-conversation extraction to preserve useful information automatically
• Scheduled consolidation so memory quality improves instead of just growing
• Dreaming and emergence workflows for longer-horizon memory refinement

• Built-in tools for web search, file operations, memory access, date/time, and more
• Advanced web search with provider chaining, fallback strategies, and live information support
• MCP integration across stdio, SSE, and Streamable HTTP transports
• Skill system with installable SKILL.md packages
• ClawHub marketplace for discovering and installing skills
• Tool guard and approval for sensitive operations

• Text-to-speech for read-aloud and voice output
• Speech-to-text for audio transcription
• Music generation
• Image generation with multiple providers
• Video generation with async task handling

Configure models in the web UI. MateClaw supports cloud and local model providers including:

• DashScope
• OpenAI
• Anthropic
• Google Gemini
• DeepSeek
• Kimi
• MiniMax
• Zhipu AI
• Volcano Engine
• OpenRouter
• Ollama
• LM Studio
• llama.cpp
• MLX

• Web app for chat, agent management, MCP, models, tools, channels, and security
• Desktop app with bundled JRE 21 and backend
• External channels for production-facing assistant workflows

Most AI systems treat knowledge like a warehouse of raw fragments.

MateClaw adds another layer: a structured Wiki that AI can build and maintain. Instead of retrieving arbitrary chunks from source files every time, the system can pre-digest information into clean pages with summaries and links.

That changes the product in three ways:

• Agents waste less context on raw material
• Knowledge becomes easier to inspect and edit by humans
• Understanding improves over time instead of resetting on every query

This is the difference between storing information and shaping it.

• Java 17+
• Node.js 18+ and pnpm
• Maven 3.9+ (or use mvnw)
• At least one LLM API key such as DashScope

Backend

cd mateclaw-server
export DASHSCOPE_API_KEY=your-key-here
mvn spring-boot:run

Backend:

• App: http://localhost:18088
• H2 Console: http://localhost:18088/h2-console
• Swagger UI: http://localhost:18088/swagger-ui.html

Frontend

cd mateclaw-ui
pnpm install
pnpm dev

Frontend:

• App: http://localhost:5173

Login

• Username: admin
• Password: admin123

cp .env.example .env
docker compose up -d

Default service:

• http://localhost:18080

Download installers from GitHub Releases.

The desktop app bundles JRE 21 + the Spring Boot backend, so users do not need to install Java separately.

macOS: if the app is blocked on first launch, use right-click → Open, or allow it in Privacy & Security.

mateclaw/
├── mateclaw-server/     Spring Boot backend
├── mateclaw-ui/         Vue 3 SPA frontend
├── mateclaw-desktop/    Electron desktop app
├── docs/                VitePress documentation
├── docker-compose.yml
└── .env.example

Backend domains include:

• agent/ for runtime and orchestration
• tool/ for built-in tools and MCP integration
• skill/ for skill installation and execution
• memory/ for extraction, consolidation, and dreaming
• wiki/ for knowledge base and structured Wiki processing
• channel/ for external platform adapters
• workspace/ for files, messages, and conversations

Current focus areas include:

• richer multi-agent collaboration
• deeper multimodal understanding
• smarter model routing
• stronger long-term memory
• richer ClawHub ecosystem
• more channels and desktop coverage

MateClaw is open to product, code, docs, and integration contributions.

git clone https://github.com/matevip/mateclaw.git
cd mateclaw

cd mateclaw-server
mvn clean compile

cd ../mateclaw-ui
pnpm install
pnpm dev

Mate means companion.
Claw means capability.

The product is meant to feel like both: a system that stays with you, and a system that can actually grab work and move it.

MateClaw is released under the Apache License 2.0.