How to run Hermes Agent on a server
Hermes is most useful when it's always reachable. That means a machine that stays on. Here's how to set one up with the official commands, and what to watch for.
Published
Hermes Agent, the open-source agent from Nous Research, keeps memory and skills and talks to you over Telegram, Discord, Slack, WhatsApp, Signal or email. On a laptop it goes quiet when the lid closes. On a server it answers at 3 a.m. Commands below are from the official docs as of October 2026.
What you need
- A Linux machine that stays on: a VPS, a home server, or a cloud computer. The installer also runs on macOS and WSL2.
- Git, curl and tar on that machine. The installer brings its own Python, Node, ripgrep and FFmpeg.
- A model provider with a model that has at least 64k tokens of context: Nous Portal, OpenAI (including a ChatGPT or Codex subscription), Anthropic, OpenRouter, Google AI Studio, a local model and others.
- Optional: a Telegram account, to talk to it from your phone.
1. Install Hermes
Sign in to the server as a normal user, not root, and run the official installer:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
source ~/.bashrc
hermes doctor
By default it also installs a Chromium browser and computer-use support. On a small server that only needs chat and the terminal, you can skip them with the installer’s --skip-browser and --skip-computer-use options. Hermes keeps everything it knows in ~/.hermes.
2. Choose a model
hermes setup
The wizard asks for a provider and credentials. To change them later, run hermes model. With a ChatGPT or Codex subscription, choose that option and sign in with the device code it shows. Keys go in ~/.hermes/.env and settings in ~/.hermes/config.yaml.
3. Connect Telegram
- In Telegram, message @BotFather, send
/newbot, and keep the token it gives you. - Get your numeric user ID from @userinfobot.
- Run the gateway setup and choose Telegram:
hermes gateway setup
Or put both values in ~/.hermes/.env:
TELEGRAM_BOT_TOKEN=<token from BotFather>
TELEGRAM_ALLOWED_USERS=<your numeric user ID>
Then start it with hermes gateway and message your bot. Users who aren’t on the allowlist are refused.
4. Keep it running
hermes gateway runs in the foreground and stops when your SSH session ends. Install it as a service instead:
# a user service; keep it running after you log out
hermes gateway install
sudo loginctl enable-linger $USER
# or, on a VPS that should bring it back at boot (Linux)
sudo hermes gateway install --system
hermes gateway status
The docs recommend the system service for a VPS or headless host. In a container, run hermes gateway run as the main process with a restart policy instead.
5. Lock it down
An agent that answers messages and runs commands deserves care. From the security docs:
- Keep the allowlist. Never set
GATEWAY_ALLOW_ALL_USERS=trueon a server. Approve new people withhermes pairing approve. - Run it as a non-root user.
- Leave approvals on. The default
smartmode asks before risky commands;--yoloturns that off. - Isolate the commands it runs. The Docker terminal backend runs commands in a locked-down container instead of on the server itself.
- Watch the logs in
~/.hermes/logs/.
6. Back up and update
hermes backup # an archive of ~/.hermes, credentials included
hermes update --backup # update, taking a backup first
Restore on a new machine with hermes import <backup.zip>. The backup contains your keys, so store it like a password.
Don’t want a server? Two other options
- Hermes Cloud. Nous runs it for you in its own container with a persistent workspace, through a Nous Portal plan, billed per second while it runs plus storage. See Nous Portal.
- A cloud computer. Your own machine, without the server administration. That’s what Cube is.
Hermes on Cube
Cube is a Linux computer in the cloud that stays on when your laptop closes. Hermes Agent is one of the 22 agents you can install in one click.
- Open your computer, then Agents & Tools, and install Hermes Agent. Cube runs the official installer.
- In a terminal on the computer, run
hermes setupand choose your model. A ChatGPT or Codex subscription works. - Set up Telegram with
hermes gateway setup, then starthermes gatewayin its own terminal. Cube’s terminals keep running after you close the app or your laptop, so the gateway stays up.
What to know:
- There’s no sudo on Cube computers yet, so the system-service install from step 4 isn’t available. Running the gateway in a Cube terminal does the same job.
- If you pause the computer, or credit runs out, running processes stop and your files stay. After Resume, start
hermes gatewayagain. Your memory, skills and settings in~/.hermesare still there. - Back up with
hermes backup, and download your files from Machine when you want a copy on your Mac.
The smallest computer costs $12 a month running all the time, and new accounts start with $25 of free credit.
Questions
Should Hermes Agent run on a Mac mini or a VPS?
Both work. A Mac mini at home is fast and you own it, but power cuts, router restarts and OS updates take your agent offline, often while you're away. A VPS or cloud computer stays reachable, at a monthly cost, and you or the provider look after it.
How big a server does Hermes Agent need?
The Hermes docs don't list minimum CPU or RAM. The model runs at your provider, so Hermes itself is light. What needs memory is everything it runs: the browser it installs for web tasks, builds and tools. The one stated requirement is a model with at least 64k tokens of context.
Can Hermes Agent use my ChatGPT or Claude subscription?
A ChatGPT or Codex subscription works: run hermes model and choose the ChatGPT or Codex subscription option, which signs in with a device code. Claude sign-in works only on a Max plan with extra usage credits; otherwise use an Anthropic API key, billed per token.