MateClaw

MateClaw is an open-source project developed by the matevip team, built upon Spring AI Alibaba. Its purpose is to enable team-level AI applications to run directly on users' own machines. The current version is 2.2.

What it aims to address is the gap that arises when an AI Agent evolves from a personal tool into a tool that enhances team productivity: namely, issues related to engineering, security, and governance. The core design principle involves decoupling the “employee” identity from the reasoning engine. Both Native and DeepSeek Harness runtime environments share the same set of mechanisms for managing sessions, workspaces, tools, and event flows. Moreover, all external processes still operate within the established boundaries of permissions and auditing controls.

A persistent goal of this approach is its relatively rare capability to break long-term tasks into manageable segments, which are then scheduled by the database queue and the supervisor. Checklists, lease agreements, attempt records, as well as any new inputs received during execution are all stored in the database, ensuring that the process can continue seamlessly after the backend is restarted. The A2A protocol is used to integrate agents from different systems under a unified governance framework.

In addition, it integrates a structured knowledge base based on the LLM Wiki framework, a multi-layer memory system, and an MCP tool ecosystem. It can be deployed via Docker, JAR files, or as a desktop application. A single JAR file is responsible for managing the runtime environment, data processing, and external communication functions. This solution is ideal for teams that have their own deployment requirements and strictly insist on keeping data within the internal network; however, individual users may find it somewhat cumbersome to use.