OpenClaw vs Hermes Agent: setup, skills and gateways

This comparison examines OpenClaw and Hermes Agent as self-hosted assistants, with deployment on separate Hostinger servers and a repeated video research brief task. The speaker favors Hermes for ongoing personal use and OpenClaw for more complex gateway routing.

The main distinction is how each handles reusable workflows. OpenClaw centers on browsing and installing skills from ClawHub; users can also ask it to create custom skills. Hermes uses a learning loop that saves workflows as skills. Its Curator process moves agent-created skills between active, stale and archived states. In the demonstrated follow-up conversation, Hermes loads a research brief skill, while OpenClaw searches memory. The speaker describes Hermes as more efficient, but provides no measured cost comparison.

The setup uses Hostinger's one-click deployment and the speaker's own model-provider API keys. The speaker discloses a Hostinger partnership. Hermes does not include a UI by default, according to the walkthrough; this deployment supplies one. OpenClaw exposes more configuration controls, which the speaker prefers for routing several messaging gateways to different agents with shared memory.

Both can run on personal hardware. The speaker recommends a disposable VPS to limit exposure of personal files and networks, and discusses risks from external skills and exposed instances.