Mateclaw
🤖 MateClaw — Your second brain with Multi-Agent Orchestration, MCP Protocol, Skills & Memory, Dream, and Multi-Channel Support. Built on Spring AI Alibaba.
Last verified:
What is Mateclaw?
MateClaw is an open-source self-hosted multi-agent AI operating system built on Spring AI Alibaba. It provides a complete team-oriented AI platform where you hire 'digital employees' with defined roles, goals, and backstories. Unlike single-user chatbots, MateClaw is designed for teams with multi-user workspaces, approval workflows for sensitive operations, complete audit logs, and production-grade health monitoring. The entire system runs from a single JAR file on your own machine, ensuring data never leaves your infrastructure.
Key features include digital employees with 5 ready-to-use职业 templates (Product Researcher, Customer Support, Knowledge Administrator, Data Analyst, Administrative Assistant), ReAct + Plan-and-Execute engines for task decomposition, LLM Wiki for transforming PDFs/Markdown/folders into structured knowledge with bidirectional linking, and a 4-layer memory system that goes from immediate context to Dreaming integration. The platform supports 15+ model providers including DashScope, Ollama, OpenAI, Kimi, and DeepSeek, with multimodal capabilities covering image generation (6 providers), video creation (4 providers), TTS, STT, and music generation.
MateClaw connects to 8 channels including Web console, desktop app (with built-in JRE 21), DingTalk, Feishu, WeChat Work, WeChat, Telegram, Discord, QQ, and Slack. The same agent with the same memory appears across all channels. The skill system uses SKILL.md protocol with MCP tool integration and ACP protocol support for bringing in top coding agents like Claude Code and Codex as employees. Tool Guard rule engine ensures all calls pass security checks with manual approval for sensitive operations.
The system is built for companies that need auditable, controllable AI deployment. It uses Apache 2.0 licensing, runs as a single process with zero hidden dependencies, and provides a Vue 3 + TypeScript console with Agent management, task lists, and multi-channel unified management. The desktop client requires no Java installation since it includes JRE 21.
Target users include development teams, enterprises needing data privacy, organizations requiring audit trails, teams wanting multi-agent collaboration, and anyone who needs AI agents that persist memory across conversations and integrate into existing workplace communication tools.
Mateclaw pricing
Pricing model: Freemium
MateClaw is completely free and open-source under Apache 2.0 license. There is no paid tier or subscription model. The software includes all features: multi-agent orchestration, MCP tools, LLM Wiki, 4-layer memory, multimodal capabilities, 8 channel integrations, and the desktop client. You only pay for the LLM API usage from your chosen 15+ model providers (DashScope, Ollama, OpenAI, Kimi, DeepSeek, OpenRouter, etc.). Ollama and OpenRouter free tier models can be used at no cost. The demo at claw-demo.mate.vip runs on their cloud for free testing with default credentials: admin/admin123.
Mateclaw pros
- Open-source Apache 2.0 license with full code auditability
- Self-hosted - data never leaves your infrastructure (0 egress)
- Single JAR file with zero hidden dependencies
- Digital employees with roles, goals, backstories and pixel art avatars
- 5 ready-to-use职业 templates out of the box
- ReAct + Plan-and-Execute engines for complex task decomposition
- LLM Wiki transforms documents into structured knowledge with bidirectional links
- 4-layer memory system from context to Dreaming integration
- Persistent PROFILE.md and MEMORY.md workspace files
- 15+ model provider support including DashScope, Ollama, OpenAI, Kimi, DeepSeek
- Multimodal built-in: 6 image + 4 video + TTS + STT + music generation
- 8 channel integration: Web, desktop, DingTalk, Feishu, WeChat Work, Telegram, Discord, Slack
- Same agent memory and personality across all channels
- Tool Guard rule engine with manual approval for sensitive operations
- ACP protocol brings Claude Code and Codex as employees
- MCP protocol for unlimited external tool expansion
- Built-in desktop client with JRE 21 - no Java installation needed
- Production-grade health monitoring and audit logs
- Multi-user workspace support for teams
- Runtime console shows what each employee is doing
- One-click employee recovery feature
- Drag-and-drop file and folder upload
- Ollama auto-discovery with 6 pre-configured models
- OpenRouter support with 200+ free tier models
Mateclaw cons
- Requires self-hosting technical knowledge for deployment
- Single JAR means all components run together - less modular
- Primarily Chinese documentation with limited English content
- Desktop app bundles JRE 21 increasing download size
- 14 built-in tools may be limited for specialized use cases
- Tool Guard approval workflow adds latency for sensitive operations
- Learning curve for SKILL.md protocol and skill creation
- Dreaming integration still experimental in early versions
- Web console requires modern browser with SSE support
- MySQL 8 database dependency increases infrastructure requirements
- Spring Boot 3 requires Java 21 minimum for custom builds
- Integration with some Chinese IM apps may require special network setup
- Memory persistence relies on workspace files that could be lost
- No cloud-hosted option available - must self-host everything
- Audit logs and health monitoring require manual configuration
Frequently asked questions about Mateclaw
What is MateClaw?
MateClaw is an open-source self-hosted multi-agent AI operating system built on Spring AI Alibaba. It provides a team-oriented AI platform where you hire 'digital employees' with roles, goals, and backstories. Unlike single-user chatbots, it offers multi-user workspaces, approval workflows, audit logs, and runs from a single JAR file on your machine with data never leaving your infrastructure.
Is MateClaw free?
Yes, MateClaw is completely free and open-source under the Apache 2.0 license. There are no paid plans or subscriptions. You only pay for LLM API usage from your chosen provider, though Ollama and OpenRouter free tier models can be used at no cost.
How do I deploy MateClaw?
MateClaw deploys as a single JAR file on your machine. Download the JAR and run it - the desktop client includes built-in JRE 21 so no Java installation is needed. For server deployment, you need Java 21 and MySQL 8. Default demo credentials are admin/admin123 at claw-demo.mate.vip.
What models does MateClaw support?
MateClaw supports 15+ model providers including DashScope, Ollama, OpenAI, Kimi, DeepSeek, and OpenRouter with 200+ free tier models. Multi-model routing allows you to use different models for different tasks with unified token usage tracking.
What channels can MateClaw connect to?
MateClaw integrates with 8 channels: Web console, desktop app, DingTalk, Feishu, WeChat Work, WeChat, Telegram, Discord, QQ, and Slack. The same agent with identical memory and personality appears across all channels simultaneously.
How does MateClaw memory work?
MateClaw uses a 4-layer memory system: immediate context during conversation, structured extraction after conversation, workspace-level PROFILE.md and MEMORY.md persistence, and定时 Dreaming that integrates scattered clues into understanding. This lets the system continue from where it left off rather than starting fresh daily.
What is the LLM Wiki feature?
LLM Wiki transforms PDFs, Markdown files, and entire folders into structured knowledge pages with bidirectional links, summaries, and source tracing. Agents automatically inject relevant knowledge. This creates a complete knowledge library you can browse and read, not fragmented vector retrieval results.
How do skills work in MateClaw?
Skills are defined by SKILL.md protocol with a manifest, tools, prompt, and LESSONS.md that makes employees smarter with use. There are 8 starter templates plus a 5-step skill creation guide. Skills include built-in, custom, and MCP types. Pre-flight checks automatically detect missing dependencies before installation.
What is Tool Guard?
Tool Guard is a rule engine that ensures all AI tool calls pass security checks. It intercepts dangerous tools and routes them through manual approval workflows, ensuring AI actions stay within control. Version v1.0.418 adds rule-based Tool Guard with审批 workflows.
Can I bring external coding agents into MateClaw?
Yes, through the ACP protocol you can bring top coding agents like Claude Code and Codex to become your digital employees. These agents integrate with the same memory, skills, and channels as native MateClaw agents.