Run Hermes on a Raspberry Pi
A Raspberry Pi is the cheapest honest way to run a self-improving agent around the clock: a one-time ~$80 board, a few watts of power, and no monthly host bill. Onepilot deploys Hermes to it over SSH and turns your iPhone into the cockpit — so the Pi can live in a drawer and you drive it from your pocket.
TL;DR
Hermes runs well on a Raspberry Pi 4 or 5 with 4 GB+ of RAM when you point it at a hosted LLM provider (the model runs in the cloud; the Pi just runs the agent loop). Install Raspberry Pi OS Lite or Ubuntu Server, put it on Ethernet, and let Onepilot handle the rest: it SSHes in, installs the Hermes CLI under nvm with no 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 Onepilot uses a real SSH tunnel — not a browser WebUI — you get the Pi's terminal, file browser, git, and cron on your phone, which matters on a headless box with no monitor.
Run a 24/7 Hermes agent on a Pi and steer it from your iPhone — one email when Onepilot ships.
Which Pi, and how much RAM?
Pi 5, 8 GB is the comfortable choice and what most people should buy. With active cooling it handles the agent loop, the SQLite/FTS5 memory Hermes keeps, and a messaging channel without breaking a sweat, and it leaves headroom if you later add a small local model for cheap tasks.
Pi 4, 4 GB is fine if the LLM lives in the cloud — which is the normal setup. Hermes itself is not memory-hungry; the heavy lifting happens at your provider. Where RAM matters is if you try to run a local model on the Pi too, which is generally not worth it on this class of hardware.
Add a decent SD card or, better, boot from USB SSD — the agent writes skill and memory files continuously, and an SSD is far kinder to that workload than a cheap card.
Headless is the whole point — and why SSH beats a WebUI here
A Pi in a drawer has no monitor and no keyboard. That is exactly the case where a browser WebUI is awkward: you would have to stand up a web server on the Pi and expose it over Tailscale or a tunnel just to see what the agent is doing. Onepilot reaches the Pi the way you already would — over SSH — so there is nothing extra to host or secure.
And on a headless box, having the real terminal in your pocket is not a luxury. When the gateway won't start, a wheel fails to build for ARM, or you need to tail a log, you run the command from the phone and fix it — no HDMI cable, no carrying the Pi to a screen.
ARM gotchas Onepilot smooths over
The classic Pi pain points are ARM Python wheels and PATH after install. Onepilot installs Hermes under nvm in your user directory, which sidesteps the system-Python mess and the sudo dance, and keeps the whole thing reversible — rm -rf ~/.nvm ~/.hermes returns the Pi to clean. If a step does fail, you see the real error string in the app (copyable), not a friendly rewrite, so you can actually act on it.
FAQ
Can a Raspberry Pi really run the Hermes agent?
Yes. A Raspberry Pi 4 or 5 with 4 GB or more of RAM runs the Hermes agent comfortably when the LLM is a hosted provider, because the model runs in the cloud and the Pi only runs the agent's control loop, its SQLite/FTS5 memory, and the messaging channel. People run Hermes on a Pi 5 8 GB as a 24/7 personal assistant for code review, Q&A, and inbox triage. Where a Pi struggles is running a large local model on the board itself, which is generally not worth it on this hardware — point Hermes at a cloud provider instead.
Do I need a monitor or keyboard for the Pi?
No, and that's the appeal. Run the Pi headless and supervise it from your phone. Onepilot connects over SSH, so you get a real terminal, file browser, git, and cron on the iPhone without ever attaching an HDMI cable. That is more practical than a browser WebUI on a headless box, because a WebUI would require you to run and expose an extra web service just to see the agent.
How much does it cost to run Hermes on a Pi?
After the one-time board purchase, the only ongoing cost is electricity — a Pi draws just a few watts, so running it 24/7 is negligible. You will still pay your LLM provider per token, but there is no host or hosting bill the way there is with a VPS. That is the core reason people choose a Pi over renting a server.
What about ARM install problems?
The usual Pi issues are missing ARM Python wheels and PATH problems after install. Onepilot installs Hermes under nvm in your user directory rather than against system Python, which avoids most of that and needs no sudo. The deploy is reversible with rm -rf ~/.nvm ~/.hermes, and if a step fails you see the actual error string in the app, copyable, so you can fix the real problem instead of guessing.
Is a Pi better than a VPS for Hermes?
It depends on where you want the box. A Pi is hardware you own with no monthly bill and full physical control, ideal for a home assistant. A VPS has no hardware to buy, better uptime and bandwidth, and is reachable from anywhere by default. Onepilot deploys Hermes the same way to both, so the choice is about ownership and location, not about the agent. See our Hermes on a VPS guide for that path.
Put a 24/7 Hermes agent on a Pi
Drop your email and we'll send one note when Onepilot ships on the App Store.
See also: Hermes on a Mac mini, Hermes on a VPS, Run Hermes on iPhone.