Your Mac Is Now an AI Agent. Here's How to Set It Up with OpenClaw

There’s a version of AI you’ve been using, and a version of AI you haven’t yet.

The version you’ve been using: you open a browser tab, type a question, get an answer, copy it somewhere, do the thing yourself.

The version you haven’t: you send a WhatsApp message from your phone while making coffee, and by the time you sit down, the task is done. Files moved. Email drafted. Research compiled. Code written. Reminders set.

That second version is what OpenClaw makes possible.

OpenClaw is a self-hosted AI agent, a gateway that runs on your Mac, connects to your LLM of choice (Claude, GPT, local models), and can actually do things on your behalf. Read and write files. Run shell commands. Search the web. Connect to Telegram, WhatsApp, or iMessage. And do all of this continuously, in the background, while you’re living your life.

This guide is for the serious setup: a fresh Mac (or a wiped one) dedicated to running OpenClaw. No distractions, no personal data risk, pure automation horsepower.

---

Why a Dedicated Machine?

Before the setup, a brief word of caution, and it’s not fearmongering.

OpenClaw runs with broad system permissions. It can read files, execute shell commands, send messages, install dependencies, and browse the web on your behalf. That’s the whole point. But it also means you genuinely don’t want it on the same machine where your iCloud Drive, passwords, and personal photos live.

The ideal OpenClaw machine is:

• A Mac Mini (M-series Apple Silicon is perfect: quiet, efficient, always-on)
• Or an old MacBook repurposed specifically for this
• Or a freshly wiped machine you’re dedicating to this purpose

If you’re reading this to set up OpenClaw on a machine you wiped clean specifically for it, you’re already thinking about this correctly.

---

What You’ll Need Before Starting

• A Mac running macOS 12 Monterey or later (Apple Silicon M1/M2/M3/M4 or Intel both work)
• At least 8GB RAM
• An Anthropic API key from console.anthropic.com (or an OpenAI key if you prefer GPT)
• A messaging app to connect (Telegram is the easiest; WhatsApp works too, just use a dedicated number)
• About 30 minutes

---

Step 1: First Boot Setup (Fresh Mac)

If you’re on a brand new or freshly wiped Mac, go through the standard macOS setup wizard. When you get to the Apple ID screen:

• Skip it if you want a truly isolated machine
• Or sign in with a secondary Apple ID you don’t use for personal data

Once you’re at the desktop, open Terminal. You’ll live here for the next 20 minutes.

Applications → Utilities → Terminal

---

Step 2: Install Homebrew

Homebrew is the package manager for macOS. Nearly every developer tool on a Mac flows through it.

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

Follow the prompts. It’ll ask for your password and may install Xcode Command Line Tools automatically, so let it. This takes about 5 minutes.

Apple Silicon users only: After Homebrew installs, run the two lines it shows at the end to add it to your PATH. They look like this:

echo 'eval "$(/opt/homebrew/bin/brew shellenv)"' >> ~/.zprofile
eval "$(/opt/homebrew/bin/brew shellenv)"

Verify it worked:

brew --version

---

Step 3: Install Node.js 22+

OpenClaw requires Node 22 or newer. Many Macs ship with older versions or none at all. Install the correct version via Homebrew:

brew install node@22

Then link it so it’s your default:

brew link node@22 --force --overwrite

Verify:

node --version   # should show v22.x.x or higher
npm --version

---

Step 4: Fix Your PATH (Common Gotcha)

The most common issue people hit after installing OpenClaw: they install it fine, try to run openclaw, and get command not found. This is almost always because npm’s global binaries directory isn’t in your shell’s PATH.

Get the correct path:

npm config get prefix

This prints something like /opt/homebrew or /usr/local. Add /bin to that and put it in your shell config:

echo 'export PATH="$(npm config get prefix)/bin:$PATH"' >> ~/.zshrc
source ~/.zshrc

Do this before installing OpenClaw. It’ll save you a headache.

---

Step 5: Install OpenClaw

The fastest path is the one-line installer. It handles everything: downloads the CLI, installs it globally via npm, and launches the onboarding wizard automatically.

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

If you’d rather install manually (you already have Node 22+ and want more control):

npm install -g openclaw@latest

Then launch onboarding:

openclaw onboard --install-daemon

The --install-daemon flag installs OpenClaw as a background service that auto-starts at boot. Highly recommended for a dedicated machine.

If you hit a sharp or libvips error (common if Homebrew installed libvips globally):

SHARP_IGNORE_GLOBAL_LIBVIPS=1 npm install -g openclaw@latest

---

Step 6: Run the Onboarding Wizard

The wizard walks you through four key things:

a) Agree to the Risk Acknowledgment

OpenClaw will ask if you understand it’s powerful and inherently risky. Use the arrow keys to select Yes. You’re on a dedicated machine, so you’ve already thought about this.

b) Choose QuickStart

Select QuickStart when prompted for installation mode. This gets you running fast.

c) Connect Your LLM

You’ll choose between Claude, GPT, or a local model. For most automation purposes:

• Claude (Anthropic): paste your API key from console.anthropic.com. You’ll be billed per usage, typically a few dollars per day for active use.
• GPT (OpenAI): paste your key from platform.openai.com
• Local model: provide your Ollama or LM Studio endpoint

d) Gateway Settings

Accept the defaults unless you have a specific reason to change ports. The default runs on port 18989.

e) Channel Setup: Connect Your Messaging App

This is the part that makes OpenClaw genuinely useful for life automation.

For Telegram (easiest):

1. Open Telegram and search for @BotFather
2. Type /newbot, give it a name and username
3. Copy the token it gives you
4. Paste it into the OpenClaw wizard

For WhatsApp: Use a dedicated prepaid SIM number, not your personal number. When OpenClaw is connected to WhatsApp, every message to it becomes agent input. You don’t want your family group chat accidentally triggering commands.

---

Step 7: Verify the Gateway Is Running

openclaw gateway status

You should see confirmation that the Gateway is active. If something’s off:

openclaw gateway start

Then open the dashboard in your browser:

openclaw dashboard

This opens the web-based Control UI at http://127.0.0.1:18789/. If it loads, you’re live. You can chat with OpenClaw directly from here without any channel setup.

If the dashboard asks for an auth token:

openclaw config get token

Paste that into the dashboard’s auth field.

---

Step 8: Set Up Auto-Start on Boot

If the --install-daemon flag worked during onboarding, this is already handled. OpenClaw manages a LaunchAgent labeled ai.openclaw.gateway that starts automatically when your Mac boots.

To verify:

launchctl list | grep openclaw

If it’s not listed, install it manually:

openclaw gateway install

Your dedicated Mac can now reboot, and OpenClaw will be back up without you touching it.

---

Step 9: Lock It Down (Security Basics)

Since this is a dedicated automation machine, take 5 minutes to reduce the attack surface:

Bind to localhost only (don’t expose the gateway to your whole network unless you need remote access):

In your openclaw.json configuration, change:

"bindAddress": "0.0.0.0"

to:

"bindAddress": "127.0.0.1"

Use Tailscale for remote access instead of opening ports to the internet. Install it:

brew install --cask tailscale

This lets you reach your OpenClaw Mac securely from your phone or other devices without exposing anything publicly.

Enable exec approval for sensitive tools in your config:

"exec_approval": true

Vet skills before installing. Community plugins can contain malware. When in doubt, run them through Cisco Skill Scanner before installing.

Set up an allowlist for your connected messaging app. Without one, OpenClaw blocks all messages by default, but once you open it up, be deliberate about who can send it commands.

---

Step 10: Put It to Work

Now the fun part. Here’s what a well-configured OpenClaw on a dedicated Mac actually lets you do:

Personal automation:

• Message it from your phone: “Check the weather in Leipzig and add an umbrella reminder if it’s going to rain”
• “Move all files from my Downloads folder older than 30 days to an archive folder”
• “Set a reminder for 6pm to call Alina”

Work automation:

• “Summarize the last 5 emails in my inbox and draft replies”
• “Run my weekly backup script and tell me if anything failed”
• “Search for recent news on BIM software updates and write me a short briefing”

Development support:

• “Check if my local server is running on port 3000 and restart it if not”
• “Pull the latest from main and run the test suite, DM me the results”

The key insight is this: you’re not using AI reactively anymore. You’re dispatching tasks to an agent that lives on a machine, has access to your tools, and reports back. That’s a fundamentally different relationship with AI.

---

Common Issues and Fixes

Problem	Fix
command not found: openclaw	Fix PATH: echo 'export PATH="$(npm config get prefix)/bin:$PATH"' >> ~/.zshrc && source ~/.zshrc
sharp install error	Run: SHARP_IGNORE_GLOBAL_LIBVIPS=1 npm install -g openclaw@latest
Dashboard auth error	Run openclaw config get token and paste into the dashboard
Gateway not starting	Run openclaw gateway install then openclaw gateway start
WhatsApp messages ignored	Check your allowlist in config. Without allowFrom, all messages are blocked
Node version error	Run brew install node@22 and brew link node@22 --force --overwrite

---

Final Thoughts

The promise of AI has always been that it handles the tedious so you can focus on the meaningful. Most tools have gotten us halfway there: answering questions faster, writing drafts, summarizing things.

OpenClaw is a step toward the other half: an AI that doesn’t just respond, but acts.

A dedicated Mac running OpenClaw 24/7 is, in a modest but real way, a digital staff member. One that doesn’t sleep, doesn’t forget, and costs you a few dollars a day in API calls rather than a salary.

That’s not nothing. Set it up properly, give it the right permissions, and keep your personal data off it, and you’ll quickly wonder how you managed workflows without it.

---

OpenClaw is open-source and self-hosted. Find the docs at docs.openclaw.ai. For macOS-specific deep dives, the official guide lives at docs.openclaw.ai/platforms/macos.

---

If this guide helped you get set up, drop a comment, and share what you end up automating first.