Skip to content

Last updated · 2026-01-30

OpenClaw Local Install (Mac / Linux)

⚠️ Security note: OpenClaw may have high-privilege capabilities.
Strongly recommended: test on a spare machine / VM / separate user account first.

What you’ll achieve

  • Install OpenClaw on Mac/Linux
  • Complete onboarding
  • Start OpenClaw and verify it works
  • Fix common issues (command not found, permissions, proxy/network, legacy conflicts)

Before you start

You’ll need

  • A terminal (Terminal on macOS, system terminal on Linux)
  • Network access to the install script (use proxy tips if needed)
  • A model API key (e.g., Anthropic/Claude)
  • ✅ Just trying it out: local is fine, ideally on a VM / spare machine
  • ✅ Long-running: use Docker/VPS for safer isolation → /docs/vps-setup

Step 0: Quick system check (optional)

bash
uname -awhoami

Step 1: Install OpenClaw (two options)

“curl | bash” is the fastest but runs remote code directly.
For beginners, it’s safer to download first and review quickly.

bash
curl -fsSL https://openclaw.bot/install.sh -o install.shsed -n '1,120p' install.shbash install.sh

Option B (fastest, higher risk)

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

Verify

bash
openclaw --version || truewhich openclaw || true

You should see a version number and a valid path (e.g., /usr/local/bin/openclaw).

Step 2: If “command not found”

This is the most common issue. Try in order:

2.1 Reopen the terminal

Many scripts update PATH. A restart often fixes it.

2.2 Reload your shell config

bash
source ~/.zshrc 2>/dev/null || truesource ~/.bashrc 2>/dev/null || true

2.3 Check PATH and install location

bash
echo $PATHwhich openclaw || true

OpenClaw was renamed from Moltbot (formerly Clawdbot). Old installs may conflict.

Symptoms

  • openclaw not found
  • Or it runs but behaves strangely

Fix

bash
npm uninstall -g clawdbot || truenpm uninstall -g moltbot || truenpm uninstall -g openclaw || true

Then re-run Step 1.

Verify

bash
which moltbot || truewhich clawdbot || truewhich openclaw || trueopenclaw --version

Step 4: Onboarding

bash
openclaw onboard --flow quickstart

You’ll likely be asked:

  • Confirm high-risk capabilities → yes
  • Pick a channel → Telegram is the easiest to verify
  • Enter your model API key
  • Choose a default model

Verify

  • Onboarding finishes without errors
  • Config saved (usually under ~/.openclaw)

Step 5: Start OpenClaw

Foreground (recommended first run):

bash
openclaw start

Background (optional):

bash
openclaw start --detach

Verify (pick one)

  1. No persistent errors in logs
  2. Send a test message in your channel and get a response

Step 6: Proxy/network issues (common in restricted networks)

Symptoms

  • Onboarding hangs
  • Skills/dependencies fail to download
  • Channel doesn’t respond

Fix: set proxy then restart

bash
export https_proxy=http://127.0.0.1:7890export http_proxy=http://127.0.0.1:7890openclaw start

Step 7: Permission errors (EACCES)

Symptoms

  • npm global install fails with EACCES
  • Writes to config/data directory fail

Recommended fixes

  • Use nvm to manage Node (avoid global install permissions)
  • Run under a user-owned directory

If you don’t want to fight permissions, use Docker/VPS: /docs/vps-setup

Common questions

Q1: Channel doesn’t respond?

  • Check network/proxy/firewall
  • Use Telegram for minimal verification
  • Read logs for error hints

Q2: Skill downloads fail (GitHub throttling)?

  • Ensure network stability
  • Use proxy in restricted networks
  • Download manually and place under the data directory (see troubleshooting)

Security reminders

  • Avoid running on your primary machine with sensitive data
  • Never commit API keys
  • Long-term use → Docker/VPS isolation → /docs/vps-setup
  • Read security hardening → /docs/advanced-config

Next steps

  • Safer deployment: /docs/vps-setup
  • Troubleshooting with screenshots: /docs/troubleshooting
  • VPS selection guide: /vps
  • Use cases and examples: /use-cases

TweetEmbed examples (if your site supports MDX components)