Hermes vs OpenClaw: VPS setup and memory explained

NetworkChuck installs Hermes Agent from Nous Research on an Ubuntu VPS and explains why he prefers it to OpenClaw after a month of use. The tutorial covers the installer, model authentication, a Telegram bot, allowed user IDs and a systemd gateway service. Hostinger sponsors the video. He runs the example as root but explicitly says this is not best practice for other setups.

Hermes is an open source AI agent harness, separate from the company's Hermes models. The model choices discussed include LM Studio for local inference and hosted providers. This self-hosted setup therefore does not require a local LLM, and the demonstrated VPS configuration uses an external model service.

Memory is a central comparison. The speaker describes limits of 1,375 characters for user.md and 2,200 for memory.md, plus periodic prompts to review what deserves retention. He also explains Honcho, an optional memory service that can run locally or in the cloud and receives messages to build context about the user.

The IT assistant demonstration connects through Twingate, creates reusable skills and controls Home Assistant devices. UniFi discovery produces another skill but needs further credentials. The Curator manages skills through active, stale and archived states. NetworkChuck's reliability claims reflect his experience rather than a guarantee. A later Kanban task stops at a usage limit, and he describes computer use as apparently still in preview.