Virix LabsHuman Browser by Virix Labs
A2A ProtocolPricingBlogCompareDocs
AccountRefundPrivacyTerms
Get Started →
June 15, 2026 · 8 min read · AI Agents · OpenClaw

OpenClaw Install Guide (2026): macOS, Linux, Windows + First Skill in 5 Minutes

A laptop, a desktop, and a server side by side with the OpenClaw logo above them and chat channel bubbles connected — OpenClaw install across platforms

If you searched openclaw install, you probably want one thing: the shortest correct path from "nothing installed" to "agent answering me from WhatsApp." This is that path. We'll cover the one-liner, the manual npm path (for people who already have Node), the Windows Hub app, and how to verify the install. Then we add the bundled browser-automation skill and pair the first chat channel. Total real time on a clean machine is about five minutes.

Disclosure: we make Human Browser, which is one of the things you can wire underneath OpenClaw's browser tool. We'll mention it where it's actually useful, not as a pitch.

Contents

  1. Requirements
  2. The one-liner install
  3. The npm path (if you already have Node)
  4. Windows specifics
  5. Verifying the install
  6. First run: openclaw onboard
  7. Adding the browser-automation skill
  8. Pairing your first channel
  9. FAQ

Requirements

Minimal, by design:

That's it. No container runtime, no separate database, no system services. The whole thing is one CLI plus a state directory under your home.

The one-liner install

On any of the three platforms, this is the cleanest install:

# macOS, Linux, Windows (PowerShell — see Windows section for native app)
$ curl -fsSL https://openclaw.ai/install.sh | bash

# The installer:
#   - installs Node.js via Homebrew (macOS) or the appropriate package
#     manager (Linux); on Windows in PowerShell it uses winget/scoop.
#   - downloads and pins the latest stable openclaw release.
#   - registers the openclaw command on your $PATH.
#
# Output ends with:
#   ✓ openclaw v0.X.Y installed
#   ▷ Next: openclaw onboard

On macOS the first run sometimes needs an Administrator password — that's Homebrew asking for permission to add itself, not OpenClaw doing anything elevated. On Linux there's no sudo prompt unless your distro requires it for the package manager step.

The npm path (if you already have Node)

If your system already has Node 22 LTS or Node 24 and you'd rather not run a shell script, skip the bash one-liner:

# verify you have a supported Node first
$ node --version
v24.0.1     # ← anything 22.19+ or 24.x is fine

$ npm i -g openclaw
$ openclaw --version

The pnpm and bun equivalents work too (pnpm add -g openclaw, bun a -g openclaw). If you intend to develop your own skills or plugins later, pnpm is the OpenClaw monorepo's native package manager and matches the upstream setup most cleanly.

For source-from-git installs (rare but supported):

$ git clone https://github.com/openclaw/openclaw.git
$ cd openclaw && corepack enable && pnpm install
$ pnpm openclaw onboard

Windows specifics

Windows users have three reasonable paths in increasing order of comfort:

  1. Windows Hub companion app — the easiest. Download the installer from openclaw.ai/install (Windows 10 20H2+ or Windows 11; x64 and ARM64 builds). It gives you a tray icon, setup wizard, chat UI, node mode, and local MCP mode without touching PowerShell.
  2. WSL2 + the Linux install — if you already live in WSL, treat it like Linux: run the bash one-liner. The browser tool docs have a dedicated "WSL2 + Windows + remote Chrome CDP" section for the case where you want the browser on the Windows host and the agent in WSL.
  3. PowerShell + native install — works, but you'll juggle path issues more than on the other two platforms. The one-liner runs under bash via Git Bash if you have it; otherwise use winget install OpenClaw.OpenClaw.
Three vertical lanes labelled macOS, Linux, and Windows with a terminal opening at the top of each and a green check mark at the bottom — cross-platform install paths

Verifying the install

One command tells you the binary works:

$ openclaw --version
openclaw v0.X.Y

$ openclaw doctor
  ✓ Node: v24.0.1
  ✓ Workdir: /home/you/.openclaw
  ✓ Gateway port: 6720 (free)
  ✓ Default model provider: not set
  ▷ Next: openclaw onboard

openclaw doctor is the "is everything okay" introspection tool. Re-run it after onboard to verify channels and skills are healthy. The --deep flag adds slower checks — model-provider ping, channel-plugin reachability, browser tool launch test. Useful for diagnosing later.

First run: openclaw onboard

The onboarding wizard is the actual setup — install puts the binary on PATH; onboard configures the Gateway. From any directory:

$ openclaw onboard

  Welcome to OpenClaw.
  We'll set up your Gateway, model provider, and first channel.

  > Model provider: [Anthropic, OpenAI, Google Gemini, Local (llama.cpp), Other]
  > Anthropic API key: sk-ant-...
  > Default model: [claude-sonnet-4-6, claude-opus-4-8, ...]
  > Workdir: /home/you/projects/my-assistant  (Enter to use this)
  > Persona: [Personal Assistant, Coder, Research, Custom]
  > Channels to enable now: [skip — you can add later]

  ✓ Gateway started on http://localhost:6720
  ✓ Web Control UI: http://localhost:6720/ui
  ✓ Welcome message sent to your default channel
  ▷ Next: openclaw skill add browser-automation

The wizard saves your choices to the workdir's .openclaw/config.json. You can re-run it any time to reconfigure — it's idempotent and doesn't reset existing skills or channels.

Install puts the binary on PATH. Onboard configures the Gateway. Skills add capabilities. Channels add inputs.

Open the Web Control UI in your browser at http://localhost:6720/ui. It's the dashboard for sessions, config, and chatting with the agent directly. Treat it as the "is this thing working" surface while you wire up real channels.

Adding the browser-automation skill

The single most useful skill to install first is the bundled browser-automation skill. Even if you don't need a browser today, having it installed means the day you do, the agent already knows the recovery loop:

$ cd /path/to/your/openclaw-workdir
$ openclaw skill add browser-automation

  ✓ added browser-automation@latest
  ▷ Skill teaches the agent: snapshot-first, stable-tab,
    stale-reference repair, manual-blocker handoff.

$ openclaw skill list
  ✓ [email protected] (bundled)

For broader pickups: openclaw skill search <topic> hits the ClawHub marketplace. Popular ones include github, linear, todoist, notion, stripe-dashboard, and dozens of community skills.

Pairing your first channel

Channels are how the agent receives input. WhatsApp and Telegram are the most popular two:

# WhatsApp
$ openclaw channel add whatsapp
  > Scan this QR code with WhatsApp → Linked Devices → Link a Device
  [QR code displayed in terminal]
  ✓ paired as "OpenClaw on iPhone — Ilya"

# Telegram
$ openclaw channel add telegram
  > Bot token (create one via @BotFather): 12345:ABCD...
  ✓ Telegram bot @YourBot ready

# verify
$ openclaw channel list
  ✓ whatsapp:   paired   (last activity: 2s ago)
  ✓ telegram:   active   (bot: @YourBot)

Send a test message — "hello" from your phone — and the agent should reply. If it doesn't, openclaw doctor --deep will pinpoint where the chain broke (model provider unreachable, channel auth expired, etc.).

The full channel matrix (Slack, Discord, iMessage, Signal, Teams, Matrix, Google Chat, Feishu, LINE, Zalo, WeChat, plus more) pairs the same way: openclaw channel add <name>. Some require platform-specific setup (a Slack app, a Discord bot token) — the CLI walks you through each.

You're running. What now?

Three good first asks for the agent:

  1. "What can you do?" — the agent walks through installed skills and channels.
  2. "Open ycombinator.com and tell me the top story." — exercises the browser-automation skill end-to-end.
  3. "Remind me at 9am tomorrow to send the deploy email." — exercises memory + standing orders.

If the second one fails because your target is protected (Cloudflare interstitial, captcha), the OpenClaw browser is hitting the data-center-IP / CDP-fingerprint wall — solvable via the remote browser profile pointed at Human Browser. The integration recipe is in our companion post: OpenClaw browser: the tool, the remote-CDP mode, and when to pair it with Human Browser.

Want stealth browser for OpenClaw protected-site tasks?

Human Browser exposes a remote CDP endpoint that drops into OpenClaw's remote browser profile. Pay-as-you-go: $0.10/hr browser time, $4/GB residential proxy, $0.005/captcha, $0.02 per agent step with the AI included. No subscription. Try it for $1 — humanbrowser.cloud.

FAQ

How do I install OpenClaw?

One-liner: curl -fsSL https://openclaw.ai/install.sh | bash. Or, if you already have Node 22 LTS / 24: npm i -g openclaw. Then openclaw onboard.

Which Node.js version?

Node 24 recommended, Node 22 LTS (22.19+) supported. Older versions are not.

Does OpenClaw run on Windows?

Yes — easiest path is the Windows Hub companion app (tray + setup + chat). Also runs in WSL2 or PowerShell.

Do I need an API key?

Yes — for at least one model provider. Anthropic, OpenAI, Gemini, or local llama.cpp. The onboard wizard asks during setup.

How do I add my first skill?

openclaw skill add browser-automation (or any other skill name). Verify with openclaw skill list.

How do I connect WhatsApp / Telegram?

openclaw channel add whatsapp (QR pairing) or openclaw channel add telegram (bot token from @BotFather). Same pattern for Slack, Discord, iMessage, Signal, and the other 15+ channels.

Related Articles

Top up — $20

Loading secure checkout…

More coins →