Descriptions:
Craig Hewitt walks through a working multi-agent setup designed to address the context-management and memory limitations common in always-on AI agent deployments. The core architecture pairs Hermes — an open-source AI agent framework — as the persistent orchestrating brain with OpenClaw instances running as specialized sub-agents, all hosted on a VPS for 24/7 availability without expensive local hardware.
The video shows the full flow live: a prompt to Hermes triggers its chief-of-staff skill, which pulls relevant context, delegates a writing task to a specialized OpenClaw sub-agent named Gary (a copywriting and email agent), collects the output, and returns it within the same unified context window. A second sub-agent, Patty, handles YouTube strategy. Hewitt explains how skills function as structured SOPs — detailed instructions that transform a blank-slate agent into a domain expert — and points to Corey Haynes’ open-source marketing skills repo as a resource for building marketing-focused agents.
The setup uses Minimax M2.7 via OpenRouter as the Hermes backbone model and is framed as entirely free to download and self-host. This video is particularly useful for practitioners frustrated by memory degradation in long-running agents who want a practical orchestration pattern that routes all context through a single persistent hub while delegating execution to cheaper, task-specific sub-agents.
📺 Source: Craig Hewitt · Published April 13, 2026
🏷️ Format: Tutorial Demo
