OpenClaw on Mac Mini: Complete Setup Guide

The Mac Mini M2 or M4 is one of the best platforms for running OpenClaw as a 24/7 home AI server — it draws only 5–10 watts at idle, runs Ollama for free local inference, and costs nothing in monthly hosting fees after the initial hardware purchase. Combined with OpenClaw's Telegram integration, you get a personal AI assistant that never sleeps.

This guide walks through the complete setup: from macOS configuration to OpenClaw installation, Ollama model selection, and Telegram bot connection.

Why Mac Mini Is Ideal for OpenClaw

Apple's M-series chips changed what's possible with local AI. The unified memory architecture means your Mac Mini can run large language models at speeds that would require a $3,000+ GPU workstation on traditional x86 hardware. A few reasons Mac Mini stands out as an OpenClaw host:

Hardware Recommendations

Choosing the right Mac Mini configuration matters. Here's the breakdown:

• Runs 7B–13B models via Ollama at 15–30 tokens/sec
• Handles OpenClaw core with 3–5 concurrent agent tasks
• 256GB is tight if you plan to store multiple models; upgrade to 512GB if possible

• Significantly faster inference than M2 at the same price point
• Runs Llama 3.1 8B at 40–55 tokens/sec
• Good for single-user power use or small team shared instance

• Runs 34B models (e.g., Qwen 2.5 34B, Mistral 22B) at usable speeds
• Multiple models loaded simultaneously without eviction
• Ideal if privacy and local-only inference is the primary goal

Step-by-Step Installation

1. Prepare macOS

Start with a fresh macOS Sequoia install or a clean user account. Enable SSH for headless access:

sudo systemsetup -setremotelogin on

Verify with ssh yourusername@mac-mini.local from another machine.

2. Install Homebrew

Homebrew is the package manager that makes the rest of the setup simple:

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

Follow the post-install instructions to add Homebrew to your PATH. For Apple Silicon, the path is /opt/homebrew/bin.

3. Install Node.js

OpenClaw requires Node.js 20 or higher. Install via nvm for easy version management:

brew install nvm
mkdir ~/.nvm
Add to ~/.zshrc:
export NVM_DIR="$HOME/.nvm"
[ -s "/opt/homebrew/opt/nvm/nvm.sh" ] && . "/opt/homebrew/opt/nvm/nvm.sh"
source ~/.zshrc nvm install 22 nvm use 22 nvm alias default 22

Verify: node --version should show v22.x.x.

4. Install OpenClaw

npm install -g openclaw
openclaw --version

Run the initial setup wizard:

openclaw init

This creates ~/.openclaw/config.json where you'll configure API keys, enabled skills, and model preferences.

5. Install and Configure Ollama

Ollama is the recommended local model runner for Mac Mini. It has native Apple Silicon optimizations and integrates cleanly with OpenClaw:

brew install ollama

Pull your first model:

ollama pull llama3.2
ollama pull qwen2.5:14b  # Recommended for 16GB+ systems

Start the Ollama server:

ollama serve

In your OpenClaw config, point to the local Ollama instance:

{
  "providers": {
    "ollama": {
      "baseUrl": "http://localhost:11434",
      "defaultModel": "qwen2.5:14b"
    }
  },
  "defaultProvider": "ollama"
}

Configuring launchd for Auto-Start

launchd is macOS's native service manager. It starts services at login, restarts them on crash, and handles logging — perfect for a headless server setup.

Create a plist for OpenClaw at ~/Library/LaunchAgents/com.openclaw.agent.plist:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN"
  "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
  <key>Label</key>
  <string>com.openclaw.agent</string>
  <key>ProgramArguments</key>
  <array>
    <string>/Users/YOURUSERNAME/.nvm/versions/node/v22.0.0/bin/openclaw</string>
    <string>serve</string>
  </array>
  <key>RunAtLoad</key>
  <true/>
  <key>KeepAlive</key>
  <true/>
  <key>StandardOutPath</key>
  <string>/Users/YOURUSERNAME/.openclaw/logs/openclaw.log</string>
  <key>StandardErrorPath</key>
  <string>/Users/YOURUSERNAME/.openclaw/logs/openclaw-error.log</string>
  <key>EnvironmentVariables</key>
  <dict>
    <key>PATH</key>
    <string>/opt/homebrew/bin:/usr/local/bin:/usr/bin:/bin</string>
  </dict>
</dict>
</plist>

Load and start:

mkdir -p ~/.openclaw/logs
launchctl load ~/Library/LaunchAgents/com.openclaw.agent.plist
launchctl start com.openclaw.agent

Create a similar plist for Ollama (com.ollama.server.plist) following the same pattern.

Disable Sleep and Configure Energy Settings

A headless server must never sleep. Open System Settings → Energy → set "Prevent automatic sleeping when the display is off" to on. Or use the command line:

# Prevent sleep
sudo pmset -a sleep 0
sudo pmset -a disksleep 0
sudo pmset -a displaysleep 10
Wake on network access (useful for WoL)
sudo pmset -a womp 1
Restart automatically after power failure
sudo pmset -a autorestart 1

Verify settings:

pmset -g

Cloudflare Tunnel for Remote Access

Instead of opening ports on your home router, Cloudflare Tunnel creates a secure outbound connection. Your OpenClaw instance becomes accessible at a stable URL with zero port forwarding.

brew install cloudflared
Authenticate
cloudflared tunnel login
Create tunnel
cloudflared tunnel create openclaw-home
Configure ~/.cloudflared/config.yml

tunnel: YOUR_TUNNEL_ID
credentials-file: /Users/YOURUSERNAME/.cloudflared/YOUR_TUNNEL_ID.json
ingress:   - hostname: openclaw.yourdomain.com     service: http://localhost:3000   - service: http_status:404

# Route DNS
cloudflared tunnel route dns openclaw-home openclaw.yourdomain.com
Run as service
sudo cloudflared service install

Now https://openclaw.yourdomain.com routes to your Mac Mini from anywhere in the world, with TLS handled automatically.

Performance Benchmarks

Real-world tokens/sec measured on Mac Mini hardware with Ollama:

Cost Analysis: Mac Mini vs VPS

• Electricity: ~$2.50/month (25W average, $0.14/kWh)
• Cloudflare Tunnel: free tier (or $0–5/month for Pro)
• Ollama: free, unlimited local inference
• Total monthly: ~$2.50–7.50
• Break-even vs $10/mo VPS: ~7 months
• Break-even vs $15/mo VPS: ~5 months

• No upfront cost
• GPU-accelerated inference for large models
• Managed infrastructure, easy scaling
• API costs for Claude/GPT add on top

• CPU-only inference: painfully slow for 7B+ models
• Fine for routing to cloud APIs, but then you're paying API costs too

When Mac Mini Beats VPS

• Privacy is non-negotiable (medical, legal, personal data never leaves your hardware)
• You want to run large local models (14B–34B) without per-token costs
• You're already paying $10+/month for VPS and it's running 24/7 anyway
• You want a persistent agent that can interact with local files and applications
• You're in a location with cheap electricity

• You need instant scalability (burst to 10x load for a day)
• You're running multiple services that already justify a VPS
• You want managed backups, snapshots, and datacenter reliability
• You don't want to manage physical hardware
• You need 99.9%+ uptime SLA (home internet isn't that reliable)