Introduction
MateClaw in One Sentence
MateClaw is a personal AI operating system for building assistants and agents that can reason, use tools, keep memory, digest knowledge, and work across real delivery surfaces.
It is not designed as a single chat screen. It is designed as a complete working system.
The Product Idea
Most AI products stop at one layer:
- a chat UI
- a tool runtime
- a memory layer
- a knowledge base
- a channel connector
MateClaw is built around the idea that these should not be separate products stitched together after the fact.
The system is designed to combine:
- Direct interaction through web and desktop
- Agent execution through reasoning loops and tool use
- Compounding memory instead of reset-on-every-chat behavior
- Structured knowledge through LLM Wiki pages, not just raw file retrieval
- Deployment surfaces such as chat, desktop, and external channels
That is the real shape of the product.
Who It Is For
MateClaw is for teams and builders who want:
- control over data and deployment
- agents that can actually perform work
- extensibility through tools, skills, and MCP
- a system that supports both development and real usage
- an AI product that grows in memory and knowledge over time
If you only want a simple hosted chatbot, this is overkill.
If you want a self-owned AI work system, this is where MateClaw becomes interesting.
Product Principles
Agents should do work
The model should not stop at elegant wording. It should decompose tasks, call tools, inspect outcomes, and continue until the work is done or blocked.
Knowledge should be shaped
Raw files are necessary, but raw files are not the final interface. MateClaw includes an LLM Wiki knowledge base so source material can become structured, linked, human-readable knowledge.
Memory should compound
A useful assistant should not start over every day. MateClaw combines short-term context, extraction, workspace files, and scheduled consolidation into one memory model.
Power requires guardrails
Tools without boundaries create risk. MateClaw includes approval, path protection, and tool-guard logic so strong capability remains usable in real environments.
AI must show up where work happens
A system that only lives in one browser tab is too small. MateClaw is built to extend into desktop and messaging/work channels.
Core Building Blocks
Agent Runtime
- ReAct
- Plan-and-Execute
- runtime-loaded agent config
- multi-agent operation
- state-graph-based execution
Knowledge and Memory
- workspace memory files
- automatic post-conversation extraction
- scheduled consolidation
- dreaming and emergence flows
- LLM Wiki knowledge base
Tools and Extension
- built-in tools
- MCP server integration
- skill packages via
SKILL.md - tool guard and approval control
Multimodal Capabilities
- speech-to-text
- text-to-speech
- music generation
- image generation
- video generation
Delivery Surfaces
- web app
- desktop app
- external channels
Why the Wiki Changes the Product
Traditional AI knowledge flows often rely on raw retrieval every time a user asks a question.
MateClaw adds another layer: a structured Wiki produced by AI from source materials. Instead of only retrieving fragments from files, the system can build linked pages with summaries and references.
This matters because:
- agents spend less context re-reading raw source material
- humans can review and edit the knowledge directly
- the system improves understanding over time instead of reconstructing it from scratch
That is a product decision, not just a technical implementation detail.
Recommended Reading Path
If you are new to MateClaw, read in this order:
Architecture Summary
MateClaw uses a monolith-with-modules design:
- Backend: Spring Boot
- Frontend: Vue 3 SPA
- Desktop: Electron with bundled backend
- Docs: VitePress
Main backend domains:
agent/tool/skill/memory/wiki/channel/workspace/
For setup, go to Quick Start.