Support daily 9:00–22:00 EETSupport 9:00–22:00 EET9:00–22:00 EET
Client areaGet started
Platform
Blog / Platform

Do you need a VPS to run OpenClaw?

OpenClaw needs a Node.js process that runs around the clock, not necessarily a server you maintain. What it needs, when a VPS pays off, what to ask a host.

I
Ihor Horhul · Founder and CEO
November 7, 2026 · 5 min read

OpenClaw is an AI agent you host yourself. It answers you in WhatsApp, Telegram or Slack and can act on your behalf, so it has to be running whenever a message arrives. Most setup guides go straight to “rent a VPS”, and some rule shared hosting out altogether — Hostinger’s OpenClaw tutorial (updated 26 March 2026) says exactly that. For classic shared hosting they are right. But “not on shared hosting” and “you need a VPS” are two different claims, and the second one depends on how much of the machine you want to look after.

inSave Hosting runs OpenClaw hosting in isolated containers: OpenClaw is installed when the container is created, and there is no server for you to maintain. Whether that or a VPS suits you comes down to the checklist below.

What OpenClaw actually needs

Strip the tutorials down and the list is short:

  • A process that never stops. The Gateway is a long-running Node.js service that holds your channel connections. On Linux it installs as a systemd user service; anything that kills background processes kills the agent.
  • A current Node.js. The docs ask for Node 24.16 or newer on the 24 line, or 26.1 and newer, with Node 26 recommended — and 24.22.1 or 26.12.1 for the security fixes. Plenty of hosts ship an older version.
  • Access to a model. A provider API key (Anthropic, OpenAI and others) or an existing Claude Code or Codex login. The model runs at the provider, so the agent needs memory, not a GPU — unless you want local models.
  • A place to keep state. Configuration, credentials, memory and the workspace live in ~/.openclaw. Lose that directory and you start from scratch.
  • No public port. The Gateway listens on port 18789 and binds to loopback by default; the docs are clear that it must never be exposed without authentication.

Memory is the one number the official docs leave out. Vendors quote their own figures (Hostinger suggests 4 GB), but the real need depends on how many channels, skills and browser sessions the agent runs.

node --version             # 24.16+ or 26.1+
openclaw gateway status    # is the Gateway running, and on which port

Why classic shared hosting fails

Classic shared hosting is built for PHP sites: your code runs while the web server answers a request, and background processes are either not allowed or stopped. There is no systemd, and Node.js is missing or old. That is the environment the “no shared hosting” advice is about, and there OpenClaw cannot stay up.

Three places to run it

Your own computer. Free and quick to try: npx openclaw@latest runs the whole onboarding in one command. But the agent goes offline whenever the laptop sleeps or the home connection drops.

A VPS. Full control: root, Docker for sandboxing the agent’s tools, any system package. You also own everything under the agent — OS patches, Node.js upgrades, openclaw update, the firewall, backups of ~/.openclaw, and keeping the Gateway off the public internet.

curl -fsSL https://openclaw.ai/install.sh | bash   # install, then onboarding starts
openclaw gateway install && openclaw gateway start  # keep the Gateway running as a service
openclaw security audit --deep                      # check exposure, auth and permissions

A managed container. A host that allows long-running Node.js processes, installs OpenClaw for you and keeps the layer under it running. You give up root and Docker; in return, patching and backups are somebody else’s job.

When a VPS is the right call

  • The agent’s tools need Docker sandboxes, or system packages a managed container will not install.
  • You run several agents for different people and want them separated at the operating-system level.
  • Heavy browser automation needs more memory than an agent plan offers.
  • You are comfortable maintaining a Linux server and want to.

A VPS does not solve local models: those need a GPU and far more RAM than any agent plan.

When it is more server than you need

One personal agent on API models, a few channels, nothing exotic installed: here a VPS mostly adds work — updates, backups and security that nobody does for you. If you pick one anyway, run openclaw security audit after every configuration change and keep ~/.openclaw in a backup that lives somewhere other than that VPS.

Questions to ask any OpenClaw host

  1. Who updates OpenClaw and Node.js, and how soon after a security release?
  2. Where is ~/.openclaw backed up, and can you restore it yourself?
  3. Can the Gateway be reached from the internet, and how are its tokens rotated?
  4. Can you get a shell and run openclaw security audit?
  5. How much RAM do you get, and what happens when you outgrow it?
  6. Model usage is billed by the provider, not the host — does the price say so?

On inSave OpenClaw hosting: the platform updates OpenClaw and Node.js automatically; the workspace is in the platform backups, with self-restore; Gateway tokens rotate automatically, and device authentication is built in; you can run the audit over SSH — here is which findings matter. Our sizing: 2 GB runs a text agent on one or two channels, 4 GB leaves room for skills and several channels, 8 GB is for heavy automation, and you can move up a plan without losing the agent. There is no root or Docker; if you need a whole server, the same OpenClaw toolkit runs on a VPS or a dedicated server of your own — ask for a quote. 14-day money-back. Setup starts with the first login guide.