Skip to content
Daleel Web

AI agents & automation

How to install OpenClaw on a VPS safely

Install the open-source OpenClaw AI assistant on your own server, connect a chat app and apply the official security settings: pairing, allowlists, sandboxing and safe skills.

By Daleel Web Editorial TeamPublished 2 min read
A shield with a lock, representing a secured AI agent
On this page

OpenClaw is one of the most popular open-source personal AI assistants. You run it on your own server and talk to it from WhatsApp, Telegram, Discord, Slack and many other apps. It's MIT-licensed and maintained by the independent OpenClaw Foundation.

That reach is also why it needs care: an agent that can read your messages and run tools on your server should be locked down from day one.

Before you start

  • A Linux VPS. OpenClaw doesn't publish a RAM minimum; a small plan works for one personal assistant.
  • Node.js 24 or 26. The installer can set it up for you. The docs recommend recent Node releases for security fixes.
  • An API key from an AI model provider.
  • A non-root user to run the assistant, not your root account.

Step 1: Install OpenClaw

The official installer detects your system, installs Node.js if needed, installs OpenClaw and starts the setup wizard:

curl -fsSL https://openclaw.ai/install.sh | bash

If you prefer npm:

npm install -g openclaw@latest --allow-scripts=openclaw
openclaw onboard --install-daemon

On npm 11.15 or older, leave out --allow-scripts=openclaw.

Step 2: Check it's running

openclaw --version
openclaw doctor
openclaw gateway status

The gateway listens on port 18789, and on a regular install it binds only to the local machine (127.0.0.1). Keep it that way.

Step 3: Connect a chat app

The docs call Telegram the fastest channel to set up, since it only needs a bot token from @BotFather. WhatsApp, Discord, Slack and others are supported too.

By default, when someone you don't know messages the assistant, they get a pairing code instead of access. You approve people one by one:

openclaw pairing approve telegram <code>

Step 4: Lock it down

These are the protections that matter most, taken from the official security guidance:

  1. Don't expose the gateway. Leave it on 127.0.0.1. To open the dashboard remotely, use an SSH tunnel or Tailscale, not an open port:
ssh -L 18789:127.0.0.1:18789 user@your-server

Then open http://127.0.0.1:18789/ in your own browser.

  1. Use allowlists for who can talk to the assistant, and require a mention in group chats.
  2. Sandbox tools. The docs note that tools run directly on the host unless you configure sandboxing. Turn on the Docker sandbox before you give the assistant access to anything important.
  3. Treat incoming messages as untrusted. Anyone who can message the assistant can try to trick it. Give it only the tools it really needs.
  4. Run the built-in audit:
openclaw security audit

Step 5: Be careful with skills

Skills and plugins come from ClawHub, a public registry where anyone can upload. Automated scans run on releases, but the official advice is clear: treat third-party skills as untrusted code and read them before enabling them. You can check a skill before installing it:

openclaw skills verify @owner/skill-name
openclaw skills install @owner/skill-name

Prefer skills from well-known authors, install as few as possible, and keep them updated with openclaw skills update --all.

For a general security checklist that applies to any self-hosted agent, see securing a self-hosted AI agent. If you're still choosing a tool, our OpenClaw vs Hermes vs n8n comparison helps.

Frequently asked questions

Is OpenClaw safe to use?

The core project has sensible defaults: the gateway listens only on the local machine and unknown senders must be approved. Most risk comes from what you add: third-party skills, exposed ports and broad tool access. Follow the steps in this guide before connecting real accounts.

Does OpenClaw cost anything?

No. OpenClaw is MIT-licensed open-source software run by an independent foundation, with no paid tier. You pay for your server and the AI model you use.

Sources & references

  1. OpenClaw on GitHub
  2. OpenClaw docs: Install
  3. OpenClaw docs: Getting started
  4. OpenClaw docs: Skills

AI agents & automation

How much does it cost to run an AI agent 24/7?

The real costs of running an AI agent: the server, the AI model's API usage and the hidden extras. A simple formula to estimate your monthly bill before you start.

2 min read