Verdict
This is the workhorse setup. Intel NUC 13 (or any equivalent 32 GB i7 mini PC) running full Hermes Agent including browser automation is what most pocket AI deployments should look like in 2026. Comfortable RAM, fast NVMe, x86 means anything Linux runs.
Setup notes
Debian 12 minimal install. Docker + docker-compose. Caddy reverse proxy with Let's Encrypt for HTTPS. Tailscale for dashboard access: never expose the dashboard publicly. Keep provider keys in ~/.hermes/.env with tight permissions, or let a secrets helper inject them at startup.
Performance
No measurement of ours. A local Mistral 7B Q4 fallback (a 4.4 GB download) fits alongside the agent runtime in 32 GB. llama.cpp's Vulkan scoreboard has no entry for this chip's graphics.
What breaks
- 70B+ local LLM (RAM ceiling)
- Apple Silicon optimised tooling
Sources
The agent's defaults and settings above were read in its own documentation on 2026-10-18.
- Hermes Agent docs: configuration (secrets in .env, secrets helper)
- Hermes Agent docs: security (API keys in ~/.hermes/.env)
Want to know more
See the Hermes Agent page and the Intel NUC 13 and Geekom IT13 spec sheet.