Run Hermes on a Mac mini
A Mac mini is the quietest, most capable always-on agent host most people can buy: low idle power, near-silent, and — on Apple Silicon — fast enough to run useful local models alongside the agent. Onepilot deploys Hermes to it over SSH and gives you the whole cockpit on your iPhone, so the mini lives under the TV and you run it from anywhere.
TL;DR
macOS is a first-class Hermes target, so a Mac mini is one of the cleanest hosts there is. An M4 with 24 GB comfortably runs Hermes plus a ~13B local model via Ollama; step up to 48 GB for 30B+ models or several concurrent sessions. Onepilot SSHes in, installs the Hermes CLI under nvm without sudo, starts it as a background process, and lets you configure the external channel it answers on — Telegram, Discord, or Slack — right in the app. Because the link is a real SSH tunnel rather than a browser WebUI, you read skill files, inspect git diffs, and run commands on the mini straight from the phone.
Make a Mac mini your always-on Hermes host and drive it from iPhone — one email when Onepilot ships.
Why a Mac mini is a great agent host
It's a supported platform, not a workaround. Hermes officially supports macOS, so installation is the happy path: the official installer pulls Python, Node, ripgrep, and ffmpeg, and the agent runs as a normal background process. No WSL, no ARM-wheel roulette.
Local models are actually viable.Apple Silicon's unified memory makes the mini one of the few affordable boxes where running a local model next to the agent makes sense — route cheap tasks to a local Ollama model and expensive reasoning to a cloud provider. An M4 24 GB handles a ~13B model comfortably.
Silent and efficient. A mini idles at a few watts and makes no noise, so it disappears into a shelf and runs for months. For a 24/7 personal agent, that matters more than raw speed.
Remote access without exposing a web service
The standard safe pattern is Tailscale plus SSH: put the mini on your tailnet and reach it from the phone with no port-forwarding and nothing listening on a public address. Onepilot connects over that SSH link directly.
This is where the SSH approach beats a browser WebUI on macOS specifically: a WebUI would mean running and exposing another web server on the mini just to interact with the agent. With Onepilot there is no extra surface — you reach the host over SSH and get the terminal, the file browser for the skill files Hermes writes, real git diffs, and cron, all on the phone. And the same app supervises OpenClaw and runs Claude Code or Codex on the same mini.
One thing to set up: keep it awake, back up ~/.hermes
Two macOS-specific notes. First, stop the mini from sleeping (Energy Saver → prevent sleep, or caffeinate) so the agent stays up when no one is logged in. Second, back up ~/.hermes/ — it holds the skills and memory database, that is, the entire learned state of a self-improving agent; a Time Machine target covers it. Onepilot installs under nvm and stays reversible with rm -rf ~/.nvm ~/.hermes.
FAQ
Is a Mac mini good for running Hermes?
Yes — it's one of the best hosts available. macOS is an officially supported Hermes platform, so installation is straightforward, and Apple Silicon's unified memory means you can run a local model via Ollama alongside the agent. A Mac mini is silent, sips power, and runs 24/7 happily, which makes it ideal as an always-on personal agent server you keep at home.
How much RAM does the Mac mini need?
If the LLM is a cloud provider, even a base mini is plenty because the agent loop itself is light. If you want to run local models on the mini, an M4 with 24 GB comfortably runs Hermes plus a ~13B model; choose 48 GB for 30B-class models or several concurrent agent sessions. A common setup is to route cheap tasks to a small local model and expensive reasoning to a cloud provider.
How do I reach the Mac mini from my iPhone?
Enable Remote Login (SSH) on the mini and put it on a private network like Tailscale, then add it to Onepilot by host and credentials. Onepilot connects over SSH — no port-forwarding and no public web service. That is safer and simpler than exposing a browser WebUI, and it gives you the mini's real terminal, file browser, git, and cron on the phone.
Will the agent keep running when I'm not logged in?
Yes, as long as the mini doesn't sleep. Set Energy Saver to prevent sleep (or use caffeinate) so the background Hermes process keeps running with no one at the keyboard. Also back up ~/.hermes/, which holds the skills and memory database — the learned state of the agent — for example with Time Machine, so a disk problem doesn't erase what the agent has learned.
Can I run other agents on the same Mac mini?
Yes. Onepilot deploys OpenClaw as well as Hermes through the wizard, and because it includes a real terminal, Claude Code and Codex CLI work on the same mini once installed. One app and one SSH layer supervise every agent on the box, rather than a separate single-purpose client per framework.
Make a Mac mini your Hermes host
Drop your email and we'll send one note when Onepilot ships on the App Store.
See also: Hermes on a Raspberry Pi, Hermes on a VPS, Run Hermes on iPhone.