Skip to content

Solutions · AI agents

A VPS for OpenClaw

OpenClaw is a gateway that sits between your chat apps and an AI agent that can run commands, use a browser and read your files — so it wants a machine that is always on, reachable only by you, and running Node 24 or newer. That is a modest server: BD-8 at $12.90 a month clears the figure hosting guides give for browser automation. Or take C2-4-80 at $36.90 and it arrives with OpenClaw already installed.

Deploy BD-8Deploy C2-4-80 with OpenClaw installed

What it needs

Software from the docs, sizes from experience

OpenClaw's documentation specifies the software and the security posture. It publishes no CPU or RAM figure; these are the sizes independent hosting guides converge on.

From docs.openclaw.ai

  • Node.js 24.16 or newer, 26 recommended. The official installer sets it up on Debian and Ubuntu.
  • A machine that stays on: the gateway is a service, installed by openclaw onboard --install-daemon.
  • The gateway binds to 127.0.0.1:18789 by default. Reach it over an SSH tunnel or Tailscale; never expose it without a token.
  • Run it as a dedicated user; unknown senders get a pairing code; groups are allow-listed.
  • An API key from your model provider. No GPU unless you host the model yourself.

Sizes guides converge on (not the project's numbers)

Guide sizing for OpenClaw workloads
WorkloadvCPURAM
Gateway + one or two chat channels, API models onlyThe common case: it waits for messages and calls a hosted model.24 GB
Plus browser automation (Playwright/Chromium)A headless browser is the heaviest thing it runs.48 GB
Local models (Ollama) alongside itThat is a model-hosting question, not an OpenClaw one; see the GPU line.832 GB

The plans that fit

Four machines, priced live

The cheapest that meets the guide figure first, then the ones that arrive with OpenClaw installed. Prices are today's, per month, read from the catalogue.

  • BD-8$12.90/mo

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

    Cheapest that fits the 4 vCPU / 8 GB guide figure. Four cores, 8 GB, 100 GB SSD and 32 TB of transfer. The supplier does not run start-up scripts, so you paste one command yourself — it is the same installer.

    Deploy
  • C2-4-80$36.90/mo

    2 vCPU · 4 GB · 80 GB disk · 3 TB

    Arrives with OpenClaw installed. The 2 vCPU / 4 GB minimum, in over thirty cities, on suppliers that run our preset. Buy it, log in, onboard.

    Deploy installed
  • C4-8-160$72.90/mo

    4 vCPU · 8 GB · 160 GB disk · 4 TB

    Arrives installed, with room for a browser. 4 vCPU / 8 GB with the preset: the size for browser automation and several channels.

    Deploy installed
  • U4-8$18.90/mo

    4 vCPU · 8 GB · 100 GB disk · Unmetered

    4 vCPU / 8 GB, unmetered, Singapore. For an agent that moves files or scrapes all day. Paste the installer yourself.

    Deploy

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

The install

Five steps, whichever row you chose

With the preset, the server arrives after step 1 is already done. Without it, step 1 is one line to paste.

  1. Install. The preset runs this for a dedicated openclaw user; on a plan without the preset, log in as root, create the user and run it yourself: curl -fsSL https://openclaw.ai/install.sh | bash -s -- --no-onboard --no-prompt --verify
  2. Onboard. As that user: openclaw onboard --install-daemon. It asks for your model provider's API key and which chat app to connect, then installs the gateway as a systemd user service.
  3. Pair a channel. Message the bot from Telegram, WhatsApp or the app you chose; OpenClaw answers unknown senders with a pairing code by default, so nobody else can use it.
  4. Open the web UI through a tunnel. ssh -N -L 18789:127.0.0.1:18789 openclaw@your-server, then http://127.0.0.1:18789/. The gateway never listens on the public address.
  5. Check it. openclaw gateway status and openclaw security audit. The preset leaves all of this in /root/OPENCLAW-NEXT-STEPS.txt.

Honestly

What we would actually buy

The cheapest machine on this page clears the guide figure. Pay more only for the preset or for room to grow.

For one OpenClaw gateway

BD-8 · $12.90/mo

4 vCPU, 8 GB, 100 GB of SSD and 32 TB of transfer: above the 4 vCPU / 8 GB that guides give for browser automation, for less than the 2 vCPU / 4 GB shapes elsewhere on this page. The one thing it does not do is run our start-up preset, so the install is the single line in step 1, typed by you. If that line is the difference, C2-4-80 at $36.90 arrives with it done.

Deploy BD-8C2-4-80 with OpenClaw installed

Questions

What people ask before they buy one.

What are OpenClaw's system requirements for a VPS?
The project documents software, not hardware: Node.js 24.16 or newer (26 recommended) on Linux, macOS or Windows, an API key from the model provider you choose, and a machine that stays on, because the gateway is a long-running service. It publishes no CPU, RAM or disk figure. Independent hosting guides converge on 2 vCPU and 4 GB as the floor for a gateway with a couple of chat channels, and 4 vCPU and 8 GB once you add browser automation, which is the heaviest thing it does. The table above uses those figures and says where they come from.
Which plan would you actually buy for OpenClaw?
BD-8 at $12.90 a month: 4 vCPU, 8 GB, 100 GB of disk and 32 TB of transfer, which is above the guide figure for browser automation and the cheapest machine on this page. Its supplier does not run start-up scripts, so the install is one pasted command instead of the preset. If you would rather the server arrive with OpenClaw already installed, C2-4-80 at $36.90 is the cheapest shape our preset runs on.
What does the start-up preset install, exactly?
The official installer, from openclaw.ai/install.sh, run in its own documented non-interactive mode (--no-onboard --no-prompt --verify) as a dedicated "openclaw" user, after the base packages it needs. It enables the user's systemd session so the gateway survives logout and reboots, opens only SSH on the firewall, and leaves the next commands in /root/OPENCLAW-NEXT-STEPS.txt. It does not touch your API key or your chat channel: you run openclaw onboard --install-daemon yourself, over SSH, which takes about five minutes. The whole script is shown in the configurator before you buy and you can edit or delete it.
Is the OpenClaw web UI exposed to the internet?
No, and it should not be. OpenClaw binds its gateway to 127.0.0.1:18789 by default and its own security guide says to keep it there and reach it through an SSH tunnel or Tailscale; binding it to a LAN or the internet requires a gateway token or password. The preset follows that: the only port open on the firewall is SSH. From your laptop, ssh -N -L 18789:127.0.0.1:18789 openclaw@your-server and open http://127.0.0.1:18789/.
Which chat apps can it connect to?
The gateway connects through channel plugins to Telegram, WhatsApp, Discord, Slack, Signal, Microsoft Teams, Google Chat, Matrix, iMessage and others. By default unknown senders get a pairing code rather than an answer, and groups are allow-listed and mention-gated. Those defaults are the reason a shared, always-on VPS is a sane place for it: the trust boundary is one operator, and the machine is reachable only by you.
Do I need a GPU?
Not for OpenClaw itself. It calls whichever model provider you configure, so the machine's job is to run the gateway, your channels and any tools it uses. A GPU enters the picture only if you want to host the model locally with something like Ollama, and then you are sizing a model server, not an agent: start from the GPU line rather than this page.
How do I keep it running and updated?
openclaw onboard --install-daemon installs the gateway as a systemd user service on Linux, which restarts it after a crash and after a reboot; the preset already enabled the user session so that works when you are not logged in. openclaw gateway status tells you it is up, openclaw security audit tells you if the configuration has drifted from the safe defaults, and openclaw doctor is the first thing to run when something looks wrong. Updates are the installer again or npm install -g openclaw@latest.

Keep reading

Related