MateClaw

MateClaw is a self-hosted team AI operating system. It is described as keeping team AI on the user’s own machine and within the deployment boundary, with agent actions that are verifiable and recoverable, a self-maintaining wiki knowledge graph, and long-term memory isolated by user. The product combines agents, wiki, memory, tools, channels, and models in one core system.

Its workflow centers on goal execution, knowledge management, memory, and runtime governance. Goals use a criteria checklist for completion and can automatically continue, replan, or delegate when work gets stuck. The wiki layer supports page types, read and write permissions, a processing pipeline, self-repair of renamed, deleted, or broken links, and extraction of entity knowledge graphs with customizable entity types. Memory includes owner-based isolation and PERSONAL, TEAM, and GLOBAL visibility controls, along with injection budgets, nightly consolidation, and file limits. The runtime layer exposes model routing, skills, MCP, ACP, channels, tools, and approvals, and administrators can view system health, agent management, tool governance, channel configuration, and execution details.

MateClaw also includes a content studio for producing publishable WeChat Official Account and Xiaohongshu content. The listed flow is topic selection, collection, writing, illustration, de-AI processing, formatting, and delivery. For WeChat, the output can go into a draft box as inline HTML and the main image can be uploaded automatically; for Xiaohongshu, it packages content into at least three vertical cards with online preview. The product also mentions content calendar deduplication and compliance scanning at delivery. Channels are described as including web, desktop, embeddable WebChat, and enterprise IM services such as Feishu, DingTalk, WeChat Work, WeChat, Telegram, Discord, and Slack, with unified handling for images, voice, and files.

Delivery methods named on the page include Docker, JAR, source code, and a desktop app. It is shown running through docker compose on localhost:18080, and the desktop release is said to include JRE 21 and support Windows 10+, macOS 14+ on Apple Silicon, and Linux via AppImage. The page also refers to an open-source, auditable system and shows a default admin login for the local deployment. A separate enterprise offering called 太一 is mentioned for private deployment in government and enterprise settings, but the main product description remains centered on the self-hosted MateClaw system.