OpenClaw on serverplus: your own 24/7 AI agent

OpenClaw is an open-source AI agent you deploy on your own server and keep running 24/7. It talks to you through the messengers you already use, remembers context, can carry out tasks (files, commands, browser automation) and connects to dozens of services. It doesn't run the model itself — you bring your own API key (OpenAI, Anthropic, etc.). So it needs no GPU — OpenClaw runs happily on an ordinary serverplus VPS.

System requirements and choosing an OS

OpenClaw has no hard minimums — the agent itself is light, because the heavy lifting is done by the external model (via API). An ordinary VPS is enough; no GPU needed.

• CPU: from 1–2 vCPU.
• RAM: at least 2 GB, comfortably 4 GB — browser automation and parallel tasks like memory.
• Disk: from 20 GB SSD (Node.js and dependencies, agent memory, browser cache, logs).
• Plus: a model API key (OpenAI, Anthropic, etc.) — you pay the provider for tokens.

Which OS to choose

OpenClaw runs on any OS, but for a server running 24/7 use a current LTS Linux distribution — easier with autostart (systemd), updates and compatibility:

• Ubuntu 24.04 LTS (or 22.04 LTS) — the recommended choice: best support and easy Node.js install.
• Debian 12 — also an excellent choice.

About Node.js. You need version 22.22+ or 24.15+. On recent LTS releases it installs in a couple of commands (see "Installation"). Avoid outdated or exotic distros — more hassle with dependencies.

Important from the start. OpenClaw actually runs commands and actions on your behalf. Deploy it on a dedicated server, not mixed with production, and take the "Security" section below seriously.

1. Installation

First Node.js (example for Ubuntu/Debian via NodeSource):

curl -fsSL https://deb.nodesource.com/setup_22.x | sudo -E bash -
sudo apt-get install -y nodejs
node -v          # should be 22.22+ or 24.15+

Then OpenClaw itself and the initial-setup wizard with an autostart service:

sudo npm install -g openclaw@latest

# onboarding: choose model, API key, install service (systemd)
openclaw onboard --install-daemon

The onboard wizard walks you through: connecting the model, basic settings and installing the daemon so the agent runs in the background and comes back up after a reboot.

2. Connecting a model

OpenClaw is model-agnostic: you specify a provider and an API key. The key and model choice are set in the onboard wizard and stored in the config:

~/.openclaw/openclaw.json

• For quality and safety, use the provider's strongest current model.
• Create a dedicated API key for the agent and set spend limits at the provider — easier to control costs.

3. How to use it

Web panel (Dashboard)

The control panel — chat with the agent, settings and sessions:

openclaw dashboard
# opens at http://127.0.0.1:18789/ by default

The panel listens on localhost only — leave it that way. Don't expose port 18789 to the internet. To reach it from your computer, use an SSH tunnel (below) or Tailscale.

Via a messenger

The main way to talk to it is from your usual messenger. OpenClaw connects to Telegram, Slack, Discord, WhatsApp, Signal, Matrix, Microsoft Teams, Google Chat and more via channels. A channel is configured in openclaw.json with an allowlist of senders:

{
  channels: {
    telegram: {
      allowFrom: ["<your_id>"],
      groups: { "*": { requireMention: true } }
    }
  }
}

• allowFrom — who is even allowed to message the agent (otherwise it replies to anyone).
• requireMention — in groups, respond only to a mention.

4. Remote access to the panel

To open the panel from your laptop without exposing the port, forward it over SSH:

ssh -L 18789:127.0.0.1:18789 user@<your-server>
# then open http://127.0.0.1:18789/ in your browser

An alternative is Tailscale (a private network between your devices and the server): also without publishing the port to the open internet.

Security

OpenClaw is powerful precisely because it acts for you — and for the same reason it demands care.

• A dedicated server. Keep the agent on a separate VPS, not next to production services and data.
• Don't expose the panel. Port 18789 on localhost only; remotely via an SSH tunnel or Tailscale.
• Allowlists. allowFrom in every channel is a must, so no outsider can talk to the agent. In groups — requireMention.
• Treat model keys as secrets. A dedicated API key for the agent, spend limits at the provider, don't commit the config to git.
• Basic server hygiene. A firewall (only SSH exposed), regular updates, SSH-key login.
• Back up config and memory. Include the ~/.openclaw directory (settings and the agent's memory in Markdown) in your backups.

Updating

sudo npm install -g openclaw@latest

After updating, restart the agent service (installed by onboard --install-daemon).

FAQ

Do I need a GPU? No. OpenClaw uses an external model over the API and only orchestrates requests — an ordinary VPS is enough.

The panel on 18789 won't open. It listens on localhost — that's correct. Open it via an SSH tunnel (ssh -L 18789:127.0.0.1:18789 ...) or Tailscale, not via a public address.

The agent doesn't reply in the messenger. Check that the channel is configured and your ID is in allowFrom. In groups the agent waits for a mention by default (requireMention).

How much does it cost? OpenClaw itself is free (MIT, open-source). You pay for the VPS and for the model's tokens at your chosen provider — set limits there to avoid surprises.

Where is the agent's data and memory stored? Locally on your server, in ~/.openclaw — nothing leaves it except the requests to your chosen model.