The Mac Mini OpenClaw Setup Guide: Local Models + Cloud Fallback

It's quiet, sips power, runs macOS (which means iMessage), and Apple Silicon handles local model inference surprisingly well. By the end of this guide, you'll have a 24/7 agent that costs $5–15/month in API fees instead of $50–150.

What You Need

Hardware

Mac Mini with Apple Silicon (M1, M2, M3, or M4)
16GB RAM minimum (24GB+ recommended for larger local models)
Ethernet cable (don't rely on WiFi for 24/7 operation)
UPS recommended (~$50 battery backup prevents dirty shutdowns during power blips)

Buying Guide (March 2026)

Model	RAM	Used Price	New Price	Local Models
M1 Mac Mini	16GB	$350–450	N/A	7B models
M2 Mac Mini	16GB	$400–500	N/A	7B models
M2 Mac Mini	24GB	$500–650	N/A	13B models ⭐
M4 Mac Mini	16GB	N/A	$500–600	7B models
M4 Mac Mini	24GB	N/A	$600–700	13B models ⭐
M2 Pro	32GB	$750–900	N/A	30B models
M4 Pro	48GB	N/A	$1,400	70B models

🟢 Best budget pick: Used M1 16GB (~$400). Handles cloud APIs perfectly and can run small local models.

⭐ Best overall pick: M2 or M4 with 24GB (~$600). Runs 13B models smoothly — covers most tasks without API costs.

1 Prepare Your Mac Mini

Initial Setup

Connect via ethernet — plug directly into your router. WiFi drops kill your agent.
Enable auto-login — System Settings → Users & Groups → Auto Login → your account
Disable sleep — System Settings → Energy Saver → Prevent automatic sleeping
Enable SSH — System Settings → General → Sharing → Remote Login → ON
Set a static IP (or reserve one in your router's DHCP settings)

Enable Headless Operation

Your Mac Mini will run without a monitor, keyboard, or mouse. Prevent display-related issues:

        # Prevent sleeping when display is off

        sudo pmset -a displaysleep 0 sleep 0 disksleep 0

        # Auto-restart after power failure

        sudo pmset -a autorestart 1

Install Prerequisites

        # Install Homebrew

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

        # Install Node.js

        brew install node

        # Verify

        node --version  # Should be v20+ or v22+

2 Install OpenClaw

        # Install OpenClaw globally

        npm install -g openclaw

        # Run the setup wizard

        openclaw setup

The wizard will ask for:

AI provider — pick Anthropic (Claude) for best agent performance
API key — get one from console.anthropic.com
Default model — start with Sonnet (best price/performance)
Messaging channel — WhatsApp, Telegram, Discord, or iMessage

Configure as Service (Auto-Start on Boot)

        # Start OpenClaw gateway as a service

        openclaw gateway start

        # Verify

        openclaw gateway status

OpenClaw will now start automatically when your Mac boots and restart if it crashes.

3 Install Ollama (Local Models)

This is where the Mac Mini really shines. Local models = $0 API costs.

        # Install Ollama

        brew install ollama

        # Start Ollama service

        brew services start ollama

        # Pull a model (Qwen 2.5 7B — excellent for agent tasks)

        ollama pull qwen2.5:7b

        # Test it

        ollama run qwen2.5:7b "Hello, are you working?"

Recommended Models by RAM

Your RAM	Model	Pull Command	Quality
8GB	Qwen 2.5 3B	`ollama pull qwen2.5:3b`	Basic tasks
16GB	Qwen 2.5 7B	`ollama pull qwen2.5:7b`	Good for most tasks
24GB	Qwen 2.5 14B	`ollama pull qwen2.5:14b`	Strong reasoning
32GB	Qwen 3.5 32B	`ollama pull qwen3.5:32b`	Near-Sonnet quality
48GB+	Llama 3 70B	`ollama pull llama3:70b`	Premium, free

4 Configure Hybrid Model Routing

The magic setup: local model for easy tasks, cloud API for hard tasks. This typically cuts monthly API costs by 70–80%.

Default model: Ollama (local) — handles greetings, quick lookups, simple tasks
Complex tasks: Claude Sonnet (cloud) — reasoning, coding, long-form writing
Heavy lifting: Claude Opus (cloud) — only when you explicitly need the best

What This Looks Like in Practice

Task	Model Used	Cost
"Good morning"	Qwen 2.5 7B (local)	$0
"What's on my calendar?"	Qwen 2.5 7B (local)	$0
"Summarize this article"	Qwen 2.5 7B (local)	$0
"Write a business proposal"	Claude Sonnet (cloud)	~$0.03
"Review this contract"	Claude Sonnet (cloud)	~$0.05
"Build a pricing strategy"	Claude Opus (cloud)	~$0.15
Heartbeat checks	Qwen 2.5 7B (local)	$0

Result: 80% of messages hit the local model ($0). Only the complex 20% use cloud API. Monthly bill: $10–20 instead of $80–150.

5 Set Up Messaging Channels

📱 iMessage (Mac-Only Feature)

The killer feature that makes Mac Mini the preferred hardware. iMessage integration only works on macOS. Text your agent like you'd text a friend — it reads and sends iMessages directly through the Messages app.

💬 WhatsApp

Scan a QR code to link your WhatsApp. Your agent can then receive and respond to WhatsApp messages.

✈️ Telegram

Create a bot via @BotFather, get a token, plug it into OpenClaw config. Most reliable channel — doesn't require QR re-linking.

🎮 Discord

Create a bot in the Discord Developer Portal, add it to your server, configure the token.

6 Customize Your Agent

SOUL.md — Your Agent's Personality

        # SOUL — My AI Assistant

        ## Identity

        - Name: [Your agent's name]

        - Role: Personal assistant + [your use case]

        - Tone: [Casual / Professional / Mixed]

        ## Boundaries

        - Never: share passwords, make purchases without asking

        - Always: ask before anything irreversible

HEARTBEAT.md — What Your Agent Does When Idle

Configure idle behavior so your agent works while you sleep — research, monitoring, documentation, anything you define.

Memory Architecture

Create a memory/ folder for persistent knowledge: key facts, projects, contacts, preferences. Your agent remembers everything between conversations.

7 Security Hardening

Don't skip this. A misconfigured agent is an open door.

API key security — store in .env file, never in committed code
Firewall — enable macOS firewall (System Settings → Network → Firewall)
SSH keys — disable password auth, use SSH keys only
Router — don't expose ports to the internet unless using a tunnel (Cloudflare, Tailscale)
Updates — keep macOS, Node.js, and OpenClaw current

Monthly Cost Breakdown

Item	Monthly Cost
Electricity (~10W average)	~$1.50
Internet (existing connection)	$0 incremental
API fees (hybrid routing)	$5–20
Hardware (amortized 2 years)	$17–30

Total: $24–52/month

Compare to managed hosting ($30–40/mo) — similar cost, but you get local models, iMessage, and full control.

Troubleshooting

Agent not responding after reboot

Check that OpenClaw gateway is running: openclaw gateway status. If not: openclaw gateway restart.

WhatsApp disconnected

WhatsApp links expire periodically. Re-scan the QR code via openclaw setup or the web UI.

Ollama model slow

Check Activity Monitor for memory pressure. If yellow/red, you need a smaller model or more RAM.

High API costs despite routing

Check your model config — make sure the local model is set as default, not just available.

Mac Mini runs hot

Normal under load. Ensure good ventilation. If sustained >95°C, check for runaway processes.

Next Level: Multiple Agents

Once comfortable with one agent, you can run multiple:

Main agent — your personal assistant
Work agent — handles business tasks with different permissions
Sub-agents — spawned by your main agent for parallel work

A Mac Mini with 24GB RAM can comfortably run 2–3 agents simultaneously.

Don't Want to Do This Yourself?

ClawReady Pro Setup ($199) — We do everything in this guide for you, plus security hardening, model routing optimization, and a 30-minute walkthrough. Running in a day, not a weekend.

Book a Pro Setup ($199) →

Or skip the hardware entirely:

Lobster Condo — Dedicated Hosting from $20/mo →

A dedicated Mac Mini in our facility with 99.5%+ uptime, monitoring, and zero maintenance.