Wiki · Start here
Installation on a VPS
From an empty server to a running agent — the practical route, including the pitfalls we hit ourselves.
Why run it on a server
The agent runs on your own computer too, but it only shows its full value when it works all the time — collecting data at night, sending you a message in the morning, keeping its tools and notes at hand even when your laptop is closed. That is what a small server (VPS) is for.
The good news: an ordinary server is enough. No GPU, no big specs — the agent mostly talks to a model over the internet.
What you need
| Item | Enough | Note |
|---|---|---|
| Server (VPS) | 1–2 cores, 2 GB RAM, 20 GB disk | Linux, ideally Ubuntu 24.04 |
| Access | IP address and a password or SSH key | Arrives by e-mail after ordering |
| System | Ubuntu, Debian, Fedora, openSUSE, Alpine… | A Raspberry Pi works too |
| Model access | An account with a model provider | Solved in step 4 |
Where to get a server: in Czechia try Wedos, vpsFree or Active24 (bank transfer or card), or Contabo / Hetzner abroad (card payment, usually cheaper). Prices change — check them before ordering; we compare them in Hosting for an AI agent. Look for "VPS" or "Cloud", not "web hosting": the agent needs its own system where it can install what it needs.
1. Order the server
Pick Ubuntu 24.04 LTS (the best-tested option), the smallest plan, and add an SSH key if you can — safer than a password. You will receive the IP address and credentials.
2. Log in and prepare the system
Type the first command — put in the IP address you received by e-mail — and press Enter:
In this demo the command types itself and underneath you see what the server really prints back: the welcome message and a waiting cursor. The server runs nothing on its own, it just waits for the next command — click the window to replay the demo.
Now type the second command (paste it and press Enter again) to update the system:
Then create your own user and work as that user, not as root:
Why not root? The agent keeps its configuration, keys and notes in its home directory. Run it as root and everything lands in root's home — and any mistake runs with root rights too. The official documentation advises against it.
3. Install the agent
You start the installation with a single command — paste it into the terminal and press Enter. The installer fetches the source and brings its own dependencies (Python, Node, ripgrep, FFmpeg), so a bare server needs no preparation:
It takes about a minute. Afterwards you must reload your shell environment — until you do, the system does not know the hermes command yet. This is the most common beginner mistake, and the easiest fix:
Optionally run the environment check, which tells you what is still missing:
4. Choose a model and sign in
The agent needs a model — the brain that does the thinking. Two sensible routes:
For option B, get a key from your provider (e.g. Nous Research, OpenAI, Anthropic, Google, OpenRouter) and paste it into the approval form that opens. Choose a model with at least 64k tokens of context — smaller ones are rejected, as they could not handle the agent's tools and memory.
Not sure which providers you can pay for from Czechia and what it costs per month? We are preparing the page Choosing a model and paying from Czechia — meanwhile ask in our Discord community.
5. Your first conversation
Just type hermes — a terminal conversation opens. Give the agent tasks whose result you can easily verify:
Try something like: "Check how much disk space is used and write it as a table." or "Find every file on the server larger than 100 MB." You will see the agent run commands itself and report back.
6. Keep it running after you close the window
The last step turns the agent into a service that comes back up after a reboot. That is what the gateway is for — the interface through which the agent receives messages (Discord, Telegram, e-mail) and runs in the background:
Without root rights, enable the so-called linger so the service keeps running after you log out: sudo loginctl enable-linger $USER.
What we ran into (and the docs do not mention)
| Situation | Fix |
|---|---|
hermes: command not found right after installing | The shell environment is not loaded — run source ~/.bashrc (or log out and back in) |
The installer reports a missing libatomic1 | Happens on minimal Debian/Ubuntu images; the installer asks for sudo itself — confirm it |
| The agent refuses to start because of the model | The selected model has less than 64k tokens of context. Pick another one (hermes model) |
| Configuration changed after an update | hermes config check (what is new) and hermes config migrate (convert old values) |
| I want to move the agent to another server | hermes backup on the old one, hermes import on the new one — keys and memory included |
| Something broke | Never delete the data directory ~/.hermes/ "to fix it" — you would lose keys and notes. Repair the installation, not the data |