HiClaw Multi-Agent Collaboration

Skill by ara.so — AI Agent Skills collection.

HiClaw is an open-source collaborative multi-agent runtime platform that enables multiple AI agents to work together in Matrix rooms with full human visibility and intervention. It uses a Manager-Workers architecture where a central Manager orchestrates multiple Worker agents, all visible in a shared IM interface.

Key Concepts

• Manager-Workers Architecture: Manager agent coordinates Worker agents, eliminating need to oversee individual workers
• Matrix Protocol: Decentralized IM protocol (Tuwunel server + Element client) for agent-human collaboration
• Multi-Runtime Support: OpenClaw (deterministic), QwenPaw (Qwen models), and Hermes (autonomous coding) workers
• MinIO Shared Storage: File system for inter-agent data exchange, reducing token consumption
• Higress AI Gateway: Centralized credential management - workers use consumer tokens, real API keys stay in gateway
• Kubernetes-Native: Custom Resource Definitions (CRDs) for declarative agent management

Installation

Docker Desktop (Quick Start)