Descriptions:
Fahd Mirza introduces HiClaw, an open-source multi-agent operating system built on top of OpenClaw that addresses a fundamental limitation of single-agent setups: the inability to coordinate parallel work across complex tasks. The video demonstrates a complete local installation on Ubuntu using an Nvidia RTX 6000 GPU with 48GB VRAM, running Ollama with the GLM 4.7 Flash model — no external API keys required.
HiClaw implements a manager-worker architecture where a central manager agent coordinates multiple worker agents through Matrix chat rooms, giving users real-time visibility into every agent conversation and the ability to intervene at any point. The security model is a standout feature: worker agents never receive actual API credentials. All keys stay in the HiClaw gateway; workers hold only consumer tokens, which Mirza argues should be a baseline requirement for any production agentic deployment.
Installation is handled by a single shell script that provisions the entire stack — AI gateway, Matrix chat server, file storage, web client, and manager agent — all containerized with Docker. Mirza walks through the interactive configuration wizard, connects the system to a local Ollama endpoint, and demonstrates the manager agent completing a coding task. He notes that while local models like GLM 4.7 Flash are sufficient for simple tasks, production multi-agent coordination is better served by hosted models from OpenAI, Anthropic, or Alibaba’s Model Studio due to the reasoning demands of the manager role.
📺 Source: Fahd Mirza · Published March 19, 2026
🏷️ Format: Tutorial Demo
