mateaix/mateclaw

What it solves

MateClaw is an enterprise-ready agent runtime and management platform designed to prevent AI outages and provide a governed environment for digital employees. It solves the problem of AI tools being tied to a single model provider or a single reasoning loop, which can lead to system failure if a vendor's API goes down or a key expires.

How it works

Built on Spring Boot, the platform decouples the "digital employee" (identity, role, and goals) from the execution engine. It uses a pluggable AgentRuntimeProvider contract, allowing users to switch between a native StateGraph engine (for ReAct and Plan-and-Execute) and a managed DeepSeek Harness (DSH) runtime via JSON-RPC. To ensure reliability, it implements a provider chain with health tracking and automatic failover to the next healthy model provider in a configured priority order.

Who it’s for

It is designed for teams and IT departments that require multi-user workspaces, RBAC, full audit trails, and the same level of governance and observability that production Java services require.

Highlights

• Multi-Vendor Failover: Automatically routes requests around failing model providers using a health tracker and cooldown window.
• Pluggable Agent Runtime: Supports both native StateGraph and external DeepSeek Harness runtimes while maintaining a single conversation and policy plane.
• LLM Wiki: A knowledge digestion system that transforms raw materials into structured, linked pages with traceable citations.
• Extensible Capabilities: Integrates with MCP (Model Context Protocol), ACP (Agent Communication Protocol), and custom SKILL.md packages.
• Enterprise Governance: Includes RBAC, JWT, approval-gated sensitive actions, and a distributed Cron lock for multi-instance deployments.
• Multi-Surface Access: Accessible via Web Console, Desktop app, Webchat widget, and multiple IM channels (e.g., Telegram, Discord, Slack).
• Business Orchestration: A linear DSL for composing employees and system actions into triggerable workflows.