Skip to content

Solutions · AI agents

A VPS for Hermes Agent

Hermes Agent is Nous Research's open-source agent: a CLI and a gateway that puts it in Telegram, Discord, Slack, WhatsApp or Signal, calling whichever model provider you configure. Its own VPS guide asks for a gigabyte of RAM, so the right server is the cheap one: U1-2 at $8.90 a month has twice that. C1-2-55 at $18.90 arrives with Hermes installed.

Deploy U1-2Deploy C1-2-55 with Hermes installed

What it needs

Light, by the project's own account

The software and the security posture come from the project's docs. The one hardware figure it publishes is the RAM floor; the rest are hosting guides' figures, labelled as such.

From hermes-agent.ai and the repository

  • Linux (or macOS/WSL2), git on the PATH. The installer brings its own Python through uv.
  • One model provider: Nous Portal, OpenRouter, OpenAI, Anthropic, or a local model.
  • At least 1 GB of RAM without a local LLM, per the project's VPS guide.
  • Runs 24/7 as a service: hermes gateway install then start.
  • Dashboard on 9119 and API on 8642 stay private: loopback plus SSH or Tailscale, or put auth in front.

Sizes by workload

Sizing for Hermes Agent workloads
WorkloadvCPURAM
Chat + gateway to one messaging app, hosted modelThe project's own VPS guide: at least 1 GB RAM without a local LLM.11 GB
24/7 with cron jobs and several channelsHosting guides' figure for continuous automation.24 GB
Browser tasks and heavier tool useA headless browser is the heaviest thing it runs.48 GB
Local model inferenceA model-hosting question, not a Hermes one; see the GPU line.832 GB

The plans that fit

Four machines, priced live

Cheapest first. Prices are today's, per month, read from the catalogue.

  • U1-2$8.90/mo

    1 vCPU · 2 GB · 25 GB disk · Unmetered

    One agent, unmetered, Istanbul. 1 vCPU, 2 GB and no traffic meter. Twice the project's minimum for less than the price of lunch. Paste the installer yourself.

    Deploy
  • C1-2$15.90/mo

    1 vCPU · 2 GB · 50 GB disk · 1 TB

    One agent, in over sixty cities. The same 1 vCPU / 2 GB shape wherever you want it, from the cheapest city. Paste the installer yourself.

    Deploy
  • C1-2-55$18.90/mo

    1 vCPU · 2 GB · 55 GB disk · 2 TB

    Arrives with Hermes installed. 1 vCPU / 2 GB on a supplier that runs our start-up preset. Buy it, log in, run hermes setup.

    Deploy installed
  • BD-8$12.90/mo

    4 vCPU · 8 GB · 100 GB disk · 32 TB

    Cron automation or browser tasks. 4 vCPU, 8 GB, 100 GB of disk: above the guide figure for browser tasks, and the cheapest 8 GB machine we sell.

    Deploy

“Deploy installed” opens the configurator with our Hermes start-up script already in the start-up box, in full, before you buy. It runs once, as root, while the server is built, and only on suppliers whose platform runs start-up scripts; on the other rows the install is one pasted command.

The install

Six commands, from the project's own guide

With the preset, step 1 is done when the server arrives. Without it, step 1 is one line.

  1. Install. As a dedicated hermes user (the preset creates it), the official installer; it brings its own Python: curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
  2. Choose a model provider. hermes setup— Nous Portal, OpenRouter, OpenAI, Anthropic or a local model. Keys go into Hermes's config, never into a prompt.
  3. Check it. hermes doctor, then hermes chat -q "Reply with one sentence confirming Hermes works."
  4. Connect a chat app. hermes gateway setup for Telegram, Discord, Slack, WhatsApp or Signal.
  5. Run it 24/7. hermes gateway install then hermes gateway start; hermes gateway status confirms it survives logout and reboot.
  6. Open the dashboard through a tunnel. ssh -N -L 9119:127.0.0.1:9119 hermes@your-server, then http://127.0.0.1:9119/. The preset leaves all of this in /root/HERMES-NEXT-STEPS.txt.

Honestly

What we would actually buy

The project asks for a gigabyte. Buy the machine that has two and spend the difference on model calls.

For one Hermes agent

U1-2 · $8.90/mo

1 vCPU, 2 GB, 25 GB of SSD and no traffic meter, in Istanbul. Need it somewhere else? C1-2 is the same shape in over sixty cities from $15.90. On both, the install is the one line in step 1, typed by you; C1-2-55 at $18.90 arrives with it done. Whichever you pick, the monthly cost of the agent will mostly be the model provider, not the server.

Deploy U1-2C1-2 in another city

Questions

What people ask before they buy one.

What VPS specs does Hermes Agent need?
Less than most pages claim. The project's own VPS guide says at least 1 GB of RAM is sufficient when the model is hosted by a provider such as Nous Portal, OpenRouter, OpenAI or Anthropic, because the agent itself is a lightweight Python process that spends its time waiting for messages and API replies. Hosting guides put continuous automation with cron jobs at 2 vCPU and 4 GB, and browser-driven tasks at 4 vCPU and 8 GB. Running a model locally is a different question entirely and needs RAM or a GPU sized for the model, not for Hermes.
Which plan would you actually buy for it?
U1-2 at $8.90 a month: 1 vCPU, 2 GB, 25 GB of disk and no traffic meter, which is twice the project's stated minimum and the cheapest machine on this page. It is in Istanbul only; C1-2 is the same shape in over sixty cities from $15.90. Neither supplier runs start-up scripts, so on both you paste the official installer yourself — one line. C1-2-55 at $18.90 is the cheapest shape that arrives with Hermes already installed by our preset.
What does the start-up preset do?
It runs the official installer from the project's GitHub repository (scripts/install.sh) in its documented --non-interactive mode as a dedicated "hermes" user, after installing curl and git. The installer brings its own Python through uv, so nothing else is needed. The preset enables the user's systemd session so the gateway survives logout and reboot, opens only SSH on the firewall, and leaves the next commands in /root/HERMES-NEXT-STEPS.txt. It does not configure a model provider or a chat app: those need your keys, which you give to hermes setup and hermes gateway setup over SSH. The script is shown in full before you buy.
How do I connect Telegram, Discord, Slack or WhatsApp?
hermes gateway setup walks through each platform and stores the token in Hermes's own config, never in a prompt. Then hermes gateway install and hermes gateway start run the gateway as a service, and hermes gateway status confirms it. One gateway process serves every channel you connect. Telegram webhooks use port 8443 if you choose webhooks over polling; polling needs no open port at all.
Is the dashboard exposed to the internet?
No. The dashboard listens on port 9119 and the OpenAI-compatible API on 8642, and the project's guidance is to keep both on loopback and reach them over SSH or Tailscale, or to put authentication in front of them. The preset opens only SSH. From your laptop: ssh -N -L 9119:127.0.0.1:9119 hermes@your-server, then open http://127.0.0.1:9119/.
Do I need a GPU?
Not unless you run the model on the server. With a hosted provider the agent is a small process and the cheapest plan here is comfortable. If you want a local model, size for the model: the project's docs put a light 3B model at 8 GB of RAM and larger models at 32 GB or more, and at that point a GPU server is the right page, not this one.
How do I keep it running after I log out?
Install the gateway as a service, not a screen session: hermes gateway install then hermes gateway start. The preset already enabled the user's systemd session (linger) so a user service keeps running with nobody logged in and comes back after a reboot. journalctl --user -u hermes-gateway -f shows its log; hermes doctor is the first command to run when something looks off.

Keep reading

Related