Run Hermes on a VPS
A VPS is the no-hardware path: a Linux box in a data center that is reachable from anywhere, up around the clock, and gone the moment you delete it. It is the fastest way to a 24/7 Hermes agent if you don't want a machine at home. Onepilot deploys Hermes to it over SSH and puts the whole cockpit on your iPhone.
TL;DR
Hermes runs on a modest Linux VPS— 1 vCPU and 1 GB of RAM is enough when the LLM is a hosted provider, since the model runs at the provider, not on the server. Spin up Ubuntu 24.04 at any host you like, and Onepilot 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. The same real SSH tunnel that does the deploy also gives you the server's terminal, file browser, git, and cron on the phone — so you never need to expose a browser WebUI on a public-facing box.
Run Hermes on a VPS, reachable from anywhere, driven from your iPhone — one email when Onepilot ships.
What size VPS does Hermes need?
Less than people expect. With a cloud LLM provider, the agent loop, its SQLite/FTS5 memory, and a messaging channel fit comfortably in 1 GB of RAM on 1 vCPU. Bump to 2 GB if you run several channels, do heavy browser automation, or want comfortable headroom. You do not need a big instance unless you intend to run a local model on the server — which on shared CPU VPS hardware is usually not worth it; point Hermes at a provider instead.
Provider choice is yours — the major budget hosts all offer an ARM or x86 instance in this class, and there are free-tier options. Onepilot is host-agnostic: it cares that the box runs Linux and accepts SSH, not whose logo is on it.
Security: a public box deserves SSH, not an exposed WebUI
A VPS has a public IP, so the security model matters more than it does on a home Pi. The right posture is boring and effective: SSH key auth only, no password login, a firewall that exposes nothing but the SSH port, and unattended security updates.
This is the strongest case for the SSH approach over a browser WebUI. Running a WebUI on a public VPS means standing up another web service — one that controls a shell-capable agent — and hardening it yourself. Onepilot adds no such surface: it reaches the agent over the same SSH you already lock down, and gives you the terminal, file browser, git diffs, and cron on the phone without anything new listening on a public port.
Disposable by design
The quiet advantage of a VPS is that it is cattle, not a pet. Onepilot installs Hermes under nvm in your user directory and keeps it reversible — rm -rf ~/.nvm ~/.hermes cleans the agent, and destroying the instance cleans everything. Snapshot the box (or just back up ~/.hermes/, the skills and memory database) before big changes, and you can rebuild from scratch in minutes.
FAQ
What are the VPS requirements for Hermes?
Modest. With a hosted LLM provider, Hermes runs comfortably on 1 vCPU and 1 GB of RAM, because the model runs at the provider and the server only runs the agent's control loop, its SQLite/FTS5 memory, and the messaging channel. Move to 2 GB if you run multiple channels or heavy browser automation. You only need a large instance if you intend to run a local model on the server itself, which usually isn't worthwhile on shared-CPU VPS hardware.
Which VPS provider should I use for Hermes?
Any Linux host works — Onepilot is provider-agnostic and only needs a box that runs Linux and accepts SSH. The major budget providers all offer an instance in the 1–2 GB class, including ARM options and free tiers. Pick on the basis of region, bandwidth, and reliability; the agent behaves identically regardless of whose hardware it is.
Is it safe to run an AI agent on a public VPS?
Yes, with standard hygiene: SSH key authentication only, no password login, a firewall that exposes nothing but the SSH port, and automatic security updates. The biggest risk people add is exposing a browser WebUI on the public IP, since that's an extra web service controlling a shell-capable agent. Onepilot avoids that entirely — it reaches the agent over SSH, so there's no additional public surface to secure.
VPS or a Raspberry Pi for Hermes?
A VPS has no hardware to buy, is reachable from anywhere by default, and offers better uptime and bandwidth, in exchange for an ongoing host bill. A Raspberry Pi is hardware you own with only an electricity cost and full physical control, ideal for a home assistant. Onepilot deploys Hermes the same way to both, so choose based on whether you want a box in a data center or one in your home.
How do I supervise the VPS agent from my phone?
Onepilot connects to the VPS over SSH and gives you the server's real terminal, a syntax-highlighted file browser to read the skill files Hermes writes, a git tab with real diffs, and cron — all on the iPhone. You direct the agent through its channel (Telegram, Discord, or Slack) and use the app to inspect, fix, and schedule. The same app also supervises OpenClaw and runs Claude Code or Codex on the same VPS.
Run Hermes on a VPS from your phone
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 Mac mini, Run Hermes on iPhone.