Skip to content
openclaw-os
Guide

Install OpenClaw โ€” step by step

Starting OpenClaw is technically a matter of minutes. But between 'works' and 'works safely and team-ready' there is a fair bit of configuration. This guide walks the full path โ€” from Node setup to first productive skill.

MS
Manuel Streit
/ / 3 min read

Prerequisites

You need:

  • Node.js 24 (or at least 22.16+) โ€” best via nvm or asdf
  • An LLM API key (OpenAI, Anthropic, Mistral) or a local Ollama install
  • Admin rights on the machine (for daemon install)
  • macOS 13+ or current Linux (Ubuntu 22.04+, Debian 12+)

One-liner install

The fastest path:

npm install -g openclaw@latest
openclaw onboard --install-daemon

The second command walks onboarding: enter API keys, pair the first channel, register the daemon as launchd (macOS) or systemd (Linux). Logs in ~/.openclaw/logs/, config in ~/.openclaw/openclaw.json.

Pair a channel

Once the daemon runs, pair the first channel. Example WhatsApp:

openclaw channel add whatsapp

You get a QR code, scan it on your phone (just like WhatsApp Web), and the channel is live. Important: before going live, set requireMention: true and define an allowlist, otherwise the bot will reply on every group.

Activate first skills

OpenClaw ships 100+ skills. By default only a few are active. Enable sensible standards:

openclaw skill enable inbox-triage
openclaw skill enable weekly-report
openclaw skill enable voice-to-task

These three are the most productive entry skills. Full list: openclaw skill list --all.

Security at setup

Before going productive: harden. At minimum:

  • requireMention: true globally
  • Channel allowlist (which numbers/groups can trigger the bot)
  • Dashboard only on 127.0.0.1, never 0.0.0.0 (see CVE-2026-25253 protection)
  • API keys via Keychain, never plaintext in config
Frequent questions

Still open questions?

Write us at hello@openclaw-os.com or book a call directly. We'll take the time.

Does OpenClaw work on a Raspberry Pi?
Theoretically yes (on Pi 4 or 5), but too weak for productive multi-channel use. Recommended: Mac mini, Intel NUC or a small cloud VM.
Do I need a GPU for local models?
For Llama 3.3 70B: yes, a GPU with 48 GB+ VRAM. For 8B models: an M-series Mac with 32 GB RAM is enough. Cloud LLMs sidestep this entirely.
How big is OpenClaw itself?
The daemon bundle is ~150 MB, skills another ~50 MB. Total install under 250 MB plus the Node runtime.

Still questions about Install OpenClaw?

Honest advice: 30 minutes, no commitment.

Book a check

This site only uses technically necessary features. Analytics loads only after consent. Cal.com booking loads only when you actively open it.