Hermes Agent hosting: Hermes Cloud, a VPS, or a managed host

What Hermes needs from a computer

Hermes is light. The only sizing figures in Nous’s docs are for its container: 1 GB of memory at the least and 2 to 4 GB recommended, one CPU core at the least and two recommended, and 2 GB or more of disk that “grows with sessions/skills”. The browser is what takes the memory: “With browser tools active, allocate at least 2 GB.” It supports Linux, macOS, Windows and Docker, and for Linux the docs say that a distribution with “glibc, systemd” and the standard filesystem layout is “likely to work pretty well”.

What it has to do all day is stay up and reach out. Its gateway is “a single background process that connects to all your configured platforms”, runs your scheduled jobs and holds your sessions. For Telegram it “makes outbound requests to Telegram’s servers to fetch new updates”, so nothing on the internet needs a way in. That is why a laptop that sleeps is a poor home for it and almost any small machine that stays on is a good one.

The model is a separate bill everywhere except where a host says otherwise. From Nous’s FAQ: “Hermes Agent itself is free and open-source (MIT license)”, and “You pay only for the LLM API usage from your chosen provider.” Hermes takes API keys, a Nous Portal subscription, and some sign-ins with a plan you already have, with limits its providers page states plainly. A Claude sign-in works only on a Max plan with extra usage credits bought (“Claude Pro subscribers cannot use this path”). For a ChatGPT sign-in, which plans are eligible and how usage counts are “not currently documented”.

The options side by side

Prices are for the computer alone, as each company’s own page gave them on October 11, 2026.

Hermes Cloud, from the people who make Hermes

Hermes Cloud is Nous Research’s own hosting, and its page describes it as “your own hosted Hermes Agent on a dedicated cloud instance that stays always on, with its own workspace and dashboard.” There is no server to see. You sign in to the Nous Portal, add credit, name the agent, choose a model and a size, and open its dashboard in the browser. It connects to Telegram, Discord, Slack and email.

• Standard: $0.51 a day running, $0.03 a day stopped, with 2 GB of memory, 2 vCPUs and 10 concurrent sessions.
• Pro: $0.99 a day running, $0.03 a day stopped, with 4 GB, 4 vCPUs and 20 concurrent sessions.

Thirty days of running comes to $15.30 on Standard and $29.70 on Pro; that sum is ours, the daily prices are Nous’s. Two lines on the page decide the real bill: “Prices are per day and exclude inference and tool usage”, and deploying needs “$2 minimum in credits, or an active subscription”. Charges come out of your Nous credit once a day. Stopping pauses the agent: it keeps its files and memory and costs the storage rate.

The page does not state the disk size, whether you get a terminal on the instance, or how to back up or export one, so ask before you depend on any of those. It suits someone who wants Hermes with nothing to administer and is content to work inside Nous’s dashboard and chat apps.

A managed host from another company

Several companies sell Hermes set up and kept running. None of them is Nous Research. These six are examples, in no order of merit, each from its own page:

They also differ in what you can touch. Agent 37 gives a terminal and a file browser in the browser, and Hostinger a web interface and a terminal. Flux, xCloud and Hermify give a dashboard, and FlyHermes names chat apps and a web chat.

The table shows why the headline price says little on its own. Before paying for any of them, check five things on the host’s own page:

• The price after the first term. Two of the six start at half their renewal price, and one of those is paid upfront.
• Whether the model is included, and capped. “AI included” with no figure is a cap you find by reaching it. With your own key, the model bill is separate and you can see it.
• What the machine is. “Dedicated” means a container on a shared server at some hosts and a virtual machine of your own at others. Hermes’s own security policy calls it “a single-tenant personal agent” whose only boundary against a misbehaving model “is the operating system”, so what sits around your instance matters.
• Whether you get a terminal. An agent with a terminal installs what its work needs. Without a shell of your own you cannot add a system package it lacks or look at what it did.
• Backups, and the way out. hermes backup writes one zip of configuration, credentials, memories, skills and sessions, and hermes import restores it on another machine. Ask whether you can run it, or download its equivalent, before you need to.

A server you rent and set up yourself

Hetzner’s price list of June 15, 2026 has a 2 vCPU, 4 GB cloud server at €5.49 a month before tax in its cost-optimized line and €19.49 in its regular line; on October 11, 2026 its site marked the cost-optimized line as not available. DigitalOcean’s is $24 a month. Installing Hermes on it is one command, and the install guide walks through it and through Docker. The work is what comes after, and Nous’s docs are specific about each piece:

• A user that is not root. “Run the source installer as the intended service user”, and from the security guide, “never run the gateway as root”.
• A service, so it survives a reboot. “Use the system service on VPS or headless hosts that should come back at boot without relying on systemd linger”: sudo hermes gateway install --system. The alternative needs no root for later restarts: the user service with sudo loginctl enable-linger $USER.
• A dashboard nobody else can reach. By default it listens on the machine itself, “only accessible from your local machine, with no login required”, and it “reads and writes your .env (API keys, secrets) and can run agent commands”. The docs’ way to use it remotely is an SSH tunnel or Tailscale, “the clean option”, where “Only devices on your tailnet can reach it.” The security policy is blunter: “Do not expose the gateway or API to the public internet without VPN, Tailscale, or firewall protection.”
• Updates. hermes update moves an install “from one release to the next”, and restarts the gateway. The newest release on October 11, 2026 was v0.21.6, published October 8. Going back is harder than going forward: “A code checkout alone is not a data rollback.”
• The system around it. Operating system updates, the firewall and SSH are yours, and the browser tool needs system libraries the installer does not add. Tailscale SSH for an agent’s server covers closing the server to everyone but your own devices.

One trap if the same server runs Claude Code or Codex: when Hermes has no sign-in of its own, “it can borrow the Codex CLI’s ~/.codex/auth.json and Claude Code’s ~/.claude/.credentials.json”, and two programs refreshing one sign-in knock each other out. The docs’ advice is to “give Hermes its own login and turn adoption off” with adopt_external_logins: false.

A one-click template

Cloud providers and deployment platforms offer Hermes as a template: a server or container with Hermes already on it. It saves the install. It does not make the result managed, and two things are worth checking on any of them.

The first is the version. DigitalOcean’s Marketplace image listed Hermes 2026.9.14 on October 11, 2026, which Nous’s release list names v0.21.3, three releases behind. The image installs Hermes for a dedicated user on Ubuntu 24.04; you log in over SSH, run hermes setup, and updating from there is yours.

The second is what the template puts on the internet. One community template on Railway pins Nous’s image at a July release and serves the dashboard on a public HTTPS address behind a username and password. That is the sign-in Hermes’s dashboard docs describe as “not suitable for exposing a dashboard directly to the public internet”, and the Docker page says why the rule tightened: “An unauthenticated public dashboard was the entry point for the June 2026 MCP-config persistence campaign”. A template that asks for a dashboard password and gives you a public link deserves a second look before your keys go into it.

A computer at home

A Mac mini, a spare laptop or a home server runs Hermes for the price of its electricity, with your files and local network in reach. hermes gateway install sets it up as a background service, which the docs recommend for “laptops and dev boxes”. The catch is the one every home machine has: the gateway does nothing while the computer sleeps or is off, and on Windows the FAQ warns that under WSL “services may not survive WSL restarts or Windows idle shutdowns”. Mac mini, VPS or a managed machine weighs that choice for any agent.

A cloud computer with Hermes already installed

Between the managed host and the bare server sits a machine that arrives with Hermes on it and is yours to run. Everpod’s developer pod is a cloud computer of your own with your pick of Claude Code, Codex, OpenCode, Pi, Hermes and OpenClaw installed, reached only over your own Tailscale network, from $24 a month for 4 vCPU and 8 GB of memory. Hermes there is installed for you to run yourself: its setup and updates are yours, on your own plan or API key. It accepts no incoming connections from the open internet: no open ports and no public SSH. A server your agent starts is reachable from your devices and from nowhere else. By default the machine does not start connections to your other devices.

If you want no administering at all, Hermes Cloud or a managed host is the better fit; if you want the lowest bill and are at home on a Linux server, rent one and install Hermes yourself.