If you just unboxed an M4 Mac Mini, you aren’t looking at a desktop computer; you’re looking at the most power-efficient AI “silo” on the planet. With 38 TOPS on the Neural Engine and unified memory, it’s the perfect hardware to run Moltbot (the artist formerly known as Clawdbot) 24/7. However, like it’s counterpart CMDR Data of the USS Enterprise, there’s some stuff to watch out for. Kidding aside – giving an AI agent shell access to your hardware is a “keys to the kingdom” moment. If you don’t secure it, you’re just building a very expensive, very fast security hole. Here are the bare minimum steps you need to take to have at least a halfway sane setup.
1. The Hardware Prep
Before touching the Moltbot installer, get your environment stable. Molbot runs great with Node.js 22.
# Install Homebrew if you haven't yet
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
# Install Node 22
brew install node@222. Installation & The “Lobster” Protocol
Run the official installer. When it asks for a method, I recommend git—it makes troubleshooting and manual updates significantly easier when things get weird.
curl -fsSL https://molt.bot/install.sh | bashOnce installed, fire up the onboarding wizard if it doesn’t start for you.
moltbot onboard --install-daemonNote on Ports: Moltbot general binds to port 18789.
3. Lock It Down!
Like Lorne on a bender we should assume the LLM will eventually hallucinate a rm -rf. We build the system to survive that.
- The Sandbox User: Create a standard macOS user (e.g.,
ai_agent). Run Moltbot under this account. This ensures that even a rogue script is walled off from your primary admin files. - Tailscale Over Port Forwarding: Never open port 18789 to the public internet. Use Tailscale. It gives your Mac Mini a private IP that only your authenticated devices (like your phone) can see. Access your dashboard at
http://[Tailscale-IP]:18789. - Confirmation Gates: Edit your
~/.moltbot/config.yml. Setrequire_approval: truefor shell tasks. If the bot wants to execute code, it has to text you for permission first.
4. M4 Performance Optimization
The M4 is a beast, but don’t get sloppy with resources. Here is the efficiency breakdown:
| Component | Recommendation | Why? |
|---|---|---|
| Model Runner | Ollama / LM Studio | Native ARM64 / Metal support. |
| Quantization | 4-bit / 8-bit | Perfect balance for the M4’s memory bandwidth. |
| Power | Prevent Sleep | Enable “Prevent automatic sleeping” in Energy Saver. |
5. Maintenance Commands
Treat this like a server. Monitor your logs regularly to see what the “lobster” is up to.
moltbot status— Check if the heart is beating.moltbot logs— See the real-time thought process.moltbot update— Don’t fall behind on security patches.
Your Mac Mini is now a high-functioning digital employee and as somewhat shiny Star Fleet Officer. Treat the security with respect, and it’ll be the best assistant you’ve ever had. If you need any data automated or custom dev done or are interested in an open-source powered self-hosted data lake, checkout Alice! Comments? Let me know on LinkedIn.
