OpenClaw Mac Mini Setup: 24/7 AI Server Guide

I run 13 AI agents on a single Mac Mini M4. They post to X, write newsletters, research sponsors, monitor analytics, and reply in my community. All from a $500 box that draws less power than a lightbulb. Here is exactly how to set yours up from scratch.

If you haven't decided between Mac Mini and VPS yet, read our complete installation guide first. Already decided? Let's build.

Why Mac Mini Is Perfect for OpenClaw

Most founders try running OpenClaw on a VPS first. It works. But the Mac Mini is better for long-term use, and here is why.

15W idle power draw. The Mac Mini M4 sips electricity. At typical US rates, that is $1 to $3 per month. A VPS costs $5 to $50 per month forever. The Mac Mini pays for itself within a year.

Apple Silicon performance. The M4 chip handles browser automation, file operations, web scraping, and sub-agent coordination without breaking a sweat. OpenClaw does not run LLMs locally (it calls cloud APIs like Claude), so the heavy compute happens elsewhere. Your Mac Mini just needs to be fast at local tasks. And Apple Silicon is very fast at local tasks.

Zero fan noise. The Mac Mini M4 is fanless under normal OpenClaw loads. Stick it on a shelf, in a closet, or under your desk. You will forget it is there.

macOS ecosystem. This is the real advantage. On a Mac Mini, your agent can access Apple Reminders, Apple Notes, iMessage, Calendar, and the full macOS automation stack (AppleScript, Shortcuts). Try that on a Linux VPS. You can't.

No monthly fees. After the one-time hardware purchase ($499 to $699), your only recurring cost is the LLM API credits. Compare that to renting a VPS indefinitely.

Always-on reliability. macOS is rock solid for server-style workloads. With the right settings (covered below), a Mac Mini will run for months without a reboot. Mine has been up for 47 days straight as I write this.

For a deeper comparison of hosting options, check out our VPS installation guide.

Which Mac Mini Should You Buy for OpenClaw?

The Mac Mini M4 with 16GB RAM is the sweet spot. It handles a main agent plus several sub-agents without slowing down. Here is how the tiers break down:

Tier	Specs	Price	Best For
Minimum	Mac Mini M4, 16GB RAM, 256GB SSD	~$499	Single agent, light workloads
Recommended	Mac Mini M4, 16GB RAM, 512GB SSD	~$699	Multiple agents, transcripts, media files
Power User	Mac Mini M4 Pro, 24GB RAM, 512GB SSD	~$1,399	13+ agents, local LLMs via Ollama, heavy browser automation

Why 16GB RAM matters

OpenClaw itself uses about 200-400MB of RAM. But when your agent spawns sub-agents, runs browser automation, and handles multiple concurrent conversations, memory usage climbs. 8GB works for a single agent doing light tasks. 16GB gives you headroom for real production workloads.

Storage: 256GB vs 512GB

macOS takes about 30GB. OpenClaw and its dependencies take another 2-3GB. The rest is for your agent's workspace: memory files, downloaded transcripts, research documents, screenshots, and logs. If your agent handles podcast transcripts or downloads media files, 256GB fills up faster than you think. The $200 upgrade to 512GB is worth it.

What about the Mac Mini M2?

Still a solid choice if you find one refurbished. The M2 with 16GB RAM handles OpenClaw perfectly well. You lose some single-core speed compared to the M4, but for an always-on agent server, the difference is minimal. Refurbished M2 Mac Minis go for $350 to $450. Great budget option.

You do not need a monitor, keyboard, or mouse after initial setup. The Mac Mini runs headless, managed entirely through Telegram and SSH.

macOS Setup and Security Hardening

Plug in a monitor and keyboard for the first 15 minutes. You will not need them again after this section.

Create a dedicated user account

Never run OpenClaw on your personal account. Your agent will store API keys, access your tools, and manage files. Keep it isolated.

1. Open System Settings, then Users & Groups, then click Add User
2. Name it something like clawagent or marc
3. Make it an Administrator (needed for Homebrew and npm global installs)
4. Set a strong password and store it somewhere safe
5. Log into this new account for all remaining steps

Enable FileVault disk encryption

Go to System Settings, then Privacy & Security, then FileVault, then Turn On. This encrypts your entire disk. If someone physically takes your Mac Mini, they cannot read your agent's data, API keys, or memory files. Non-negotiable for any machine storing credentials.

Enable the macOS firewall

System Settings, then Network, then Firewall, then Turn On. Block all incoming connections except the ones you explicitly allow. OpenClaw does not need any incoming ports open. It makes outgoing connections to LLM APIs and messaging services.

Enable automatic software updates

System Settings, then General, then Software Update. Turn on automatic updates for macOS and App Store apps. Security patches should install automatically on a server you rarely touch physically.

Installing Node.js on Mac Mini

OpenClaw requires Node.js v18 or later. The cleanest way to install it is through Homebrew.

# Install Homebrew (if not already installed)
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Follow the post-install instructions to add brew to your PATH
# Then install Node.js
brew install node

# Verify the installation
node --version  # Should show v18+ (v22+ recommended)
npm --version

If you prefer a version manager (useful if you run other Node.js projects), use nvm or fnm instead:

# Using fnm (fast node manager)
brew install fnm
fnm install --lts
fnm use lts-latest

Installing OpenClaw Step by Step

The fastest path is the interactive installer at installopenclawnow.com. It detects your OS and walks you through everything.

Or install manually in three commands:

# Install OpenClaw globally
npm install -g openclaw

# Run the onboarding wizard with daemon mode
openclaw onboard --install-daemon

The onboarding wizard will walk you through:

1. Choosing your LLM provider (Anthropic recommended for Claude Opus)
2. Entering your API key
3. Setting up a messaging channel (Telegram, Discord, or WhatsApp)
4. Naming your agent
5. Creating the workspace directory at ~/.openclaw/workspace/

For the full walkthrough with screenshots, see our complete OpenClaw installation guide.

How Do You Configure API Keys for OpenClaw?

OpenClaw needs at least one LLM API key to function. The recommended setup is Anthropic's Claude API for the main model.

Getting your Anthropic API key

1. Go to console.anthropic.com
2. Create an account or sign in
3. Navigate to API Keys and create a new key
4. Add billing info and load credits ($20 is a good starting amount)
5. Copy the key (starts with sk-ant-)

We have a detailed Anthropic API key setup guide if you need help with billing tiers or usage limits.

Adding the key to OpenClaw

If you used the onboarding wizard, it already prompted you for the key. To add or update keys later:

# Open the OpenClaw config
openclaw config edit

# Or set the key directly
openclaw config set anthropic.apiKey sk-ant-your-key-here

Optional: Add keys for other services

Depending on what your agent does, you might want additional API keys:

Service	What It Enables	Required?
Anthropic (Claude)	Main LLM for agent reasoning	Yes
OpenAI	Whisper transcription, DALL-E images, GPT fallback	Optional
Brave Search	Web search from your agent	Optional (free tier available)
ElevenLabs	Text-to-speech voice responses	Optional

Store these in your TOOLS.md file or in the OpenClaw config. Your agent can reference them when needed.

Setting Up Mac Mini as a 24/7 Server

By default, macOS aggressively sleeps inactive machines. Your agent needs to run around the clock. Here is how to prevent that.

Disable all sleep modes

# Prevent system sleep, display sleep, and disk sleep
sudo pmset -a sleep 0 displaysleep 0 disksleep 0

# Enable auto-restart after power failure
sudo pmset -a autorestart 1

# Disable hibernation (not needed for a desktop)
sudo pmset -a hibernatemode 0

Verify your settings:

pmset -g

You should see sleep 0, displaysleep 0, disksleep 0, and autorestart 1.

Enable auto-login

For the Mac Mini to fully boot into your agent's user account without manual input:

1. Open System Settings, then Users & Groups
2. Click the "i" icon next to your agent user
3. Enable "Automatically log in as" and select the agent account

Disable screen lock

System Settings, then Lock Screen. Set "Require password after screen saver begins or display is turned off" to Never. On a headless server with no monitor, screen lock just creates problems.

Schedule weekly restarts (optional but recommended)

A fresh reboot once a week clears memory leaks and keeps things snappy:

# Restart every Sunday at 4:00 AM
sudo pmset repeat restart MTWRFSU 04:00:00

OpenClaw's daemon restarts automatically after reboot, so there is zero downtime impact.

Connecting Telegram to Your AI Agent

During onboarding, you will set up a messaging channel. Telegram is the most popular choice, and for good reason. It works on every device, supports rich formatting, and the bot API is free.

1. Open Telegram and search for @BotFather
2. Send /newbot and follow the prompts
3. Choose a name and username for your bot
4. Copy the bot token (looks like 123456:ABC-DEF...)
5. Paste it into the OpenClaw onboarding wizard when prompted
6. Open a chat with your new bot and send "Hello"

Your agent is now live. Message it from your phone while sitting on a beach in Bali and it responds from your Mac Mini at home. That is the setup I run every day.

Want to use Discord instead? Check our OpenClaw Discord bot guide.

How Do You Access Your Mac Mini Remotely?

Once the Mac Mini is headless (no monitor), you need reliable ways to access it from anywhere. There are three methods, and I recommend using all of them.

Method 1: SSH (command line access)

Enable Remote Login in System Settings, then General, then Sharing, then Remote Login. This lets you SSH into the Mac Mini from any other machine on the same network:

# From another Mac on the same network
ssh clawagent@mac-mini.local

# Check OpenClaw status
openclaw gateway status

# View agent logs
openclaw logs

Method 2: Screen Sharing (full desktop access)

Enable Screen Sharing in System Settings, then General, then Sharing. This gives you full graphical access to the Mac Mini desktop. Useful when you need to interact with browser windows or debug visual issues.

From another Mac, open Finder, then Go, then Connect to Server, then enter vnc://mac-mini.local.

Method 3: Tailscale (access from anywhere)

SSH and Screen Sharing work on your local network. But what if you are traveling? Tailscale creates a secure, private network between all your devices. No port forwarding. No exposing anything to the public internet.

# Install Tailscale on the Mac Mini
brew install tailscale

# Authenticate
tailscale up

Install Tailscale on your laptop and phone too. Then you can SSH from anywhere:

# SSH from a coffee shop in Tokyo
ssh clawagent@mac-mini.tail1234.ts.net

# Or use Screen Sharing from anywhere
# Connect to vnc://mac-mini.tail1234.ts.net

Configuring Your Agent's Personality

Your agent's workspace lives in ~/.openclaw/workspace/. These markdown files define everything about how your agent thinks, speaks, and operates:

File	Purpose
SOUL.md	Your agent's personality, tone, and communication style
AGENTS.md	Operating rules, boot sequence, safety constraints, workflow SOPs
USER.md	Everything about you: timezone, goals, preferences, business context
TOOLS.md	API keys, account details, tool configurations, service notes
HEARTBEAT.md	Periodic task checklist your agent runs every 30 minutes
IDENTITY.md	Your agent's name, role, and emoji

Start with a simple SOUL.md ("Be concise. No corporate speak. Help me ship.") and a USER.md with your timezone and main project. Your agent will help you iterate and expand from there.

Want production-ready templates? Read our guide to running 13 agents on a single Mac Mini. It includes the exact SOUL.md, AGENTS.md, and SOP files I use.

Can You Run Multiple Agents on One Mac Mini?

Yes. And this is where the Mac Mini setup really shines.

A single Mac Mini M4 with 16GB RAM comfortably runs a main agent plus 10 to 15 sub-agents. Each sub-agent is a specialized worker: one for X/Twitter, one for newsletter drafts, one for YouTube research, one for podcast automation, one for community management.

How multi-agent works on one machine

Your main agent is the coordinator. It spawns sub-agents as background processes when tasks come in. Sub-agents do their work, report back, and shut down. The main agent stays running 24/7.

Resource usage for a typical 13-agent setup:

Resource	Usage	M4 16GB Capacity
RAM	2-4GB active	16GB (plenty of headroom)
CPU	5-15% average	10-core M4 (barely working)
Disk	10-20GB workspace	256-512GB SSD
Network	Minimal (API calls)	Gigabit Ethernet or WiFi 6E

Structuring your agent team

Each sub-agent gets its own SOP file in the sops/ directory. The SOP defines exactly what it does, which tools it can use, what cron jobs it runs, and how it reports back to the main agent.

Example structure from my setup:

~/.openclaw/workspace/
├── SOUL.md           # Main agent personality
├── AGENTS.md         # Coordination rules
├── USER.md           # About me (Florian)
├── sops/
│   ├── dan-x-growth.md      # X/Twitter posting
│   ├── tyler-newsletter.md  # Newsletter drafts
│   ├── jimmy-youtube.md     # YouTube optimization
│   ├── bob-ops.md           # Operations & dispatch
│   ├── loop-analytics.md    # Analytics monitoring
│   └── ... (8 more agents)

Inside OpenClaw Lab, we share the complete multi-agent architecture. SOPs, cron schedules, coordination playbooks, and the exact prompts that make each agent effective. Copy-paste and adapt to your business.

Performance Tips and Real Benchmarks

After running OpenClaw on a Mac Mini M4 for months, here is what I have learned about squeezing out the best performance.

Real-world benchmarks

Task	Time on Mac Mini M4	Notes
Agent cold start	2-3 seconds	launchd to first response
Telegram message response	3-8 seconds	Depends on Claude API latency
Browser automation (full page)	4-6 seconds	Load + screenshot + parse
Sub-agent spawn	1-2 seconds	Background process creation
File search (10K files)	Under 1 second	SSD speed matters here

The bottleneck is never the Mac Mini. It is always the LLM API response time. Your local machine just needs to be fast enough to run tools, manage files, and handle browser automation. The M4 is more than enough.

Tips for better performance

Use Ethernet instead of WiFi. The Mac Mini M4 has Gigabit Ethernet built in. Plug it in. WiFi adds latency and occasional drops that interrupt agent workflows.

Close unused apps. On a dedicated agent server, nothing else should be running. No Chrome, no Spotify, no Slack. Just OpenClaw and the system processes it needs.

Monitor disk space. When the SSD gets above 80% full, macOS slows down. Set up a cron job in your agent to alert you when disk usage crosses 75%.

Use efficient models for simple tasks. Not every sub-agent needs Claude Opus. For simple formatting, file operations, or template-based work, smaller models reduce API costs and respond faster.

Keep OpenClaw updated. New versions regularly include performance improvements:

# Update OpenClaw to latest
npm update -g openclaw

# Restart the daemon
openclaw gateway restart

Mac Mini AI Server: Monthly Costs Breakdown

Item	Cost	Notes
Mac Mini M4 16GB	$499-699 (one-time)	Pays for itself vs VPS in 10-14 months
Claude Opus API credits	$20-150/month	Depends on usage volume
Electricity (~15W idle)	$1-3/month	Less than charging your phone
Tailscale	Free	Personal plan, up to 100 devices
Telegram bot	Free	No limits on bot API
Total monthly	$21-153/month	After initial hardware purchase

Compare that to hiring a virtual assistant ($500 to $2,000/month), any SaaS automation platform ($50 to $500/month), or a comparable cloud server ($20 to $100/month with no macOS features).

The Mac Mini is the cheapest way to run a 24/7 AI agent that actually does useful work.

Troubleshooting Common Mac Mini Issues

After helping hundreds of founders set up their Mac Mini servers inside OpenClaw Lab, these are the most common problems and fixes.

Mac Mini goes to sleep despite pmset settings

This usually means Power Nap is still enabled. Disable it:

sudo pmset -a powernap 0
sudo pmset -a proximitywake 0
sudo pmset -a tcpkeepalive 1

Also check that no energy-saving app (like Amphetamine) is overriding your pmset settings with its own schedule.

OpenClaw does not start after reboot

Check that the launchd daemon is loaded:

# Check daemon status
launchctl list | grep openclaw

# If not listed, reload it
openclaw onboard --install-daemon

Make sure auto-login is enabled for the correct user account. The daemon runs in the user context, so the user must be logged in.

SSH connection refused

Verify Remote Login is enabled: System Settings, then General, then Sharing, then Remote Login. Make sure "Allow access for" includes your agent user account or is set to All Users.

High memory usage over time

Long-running browser automation sessions can accumulate memory. The weekly restart schedule mentioned earlier handles this. If you see issues between restarts:

# Check what is using memory
top -l 1 -o mem | head -20

# Restart just the OpenClaw gateway
openclaw gateway restart

Agent stops responding but Mac Mini is online

Usually an API key issue (expired credits) or a network blip. SSH in and check:

# Check gateway status
openclaw gateway status

# Check recent logs for errors
openclaw logs --tail 50

Nine times out of ten, the fix is topping up your Anthropic API credits.

Unplug the Monitor and Ship

You are done with physical access. Disconnect the monitor, keyboard, and mouse. Tuck the Mac Mini on a shelf or behind your router. It draws about 15W idle. Less than a lightbulb. Quieter than a sleeping cat.

From here, everything happens through Telegram. Tell your agent what to do. It does it. 24 hours a day, 7 days a week, from a box that costs less per month than your coffee habit.

That is the setup. Now go build something with it.

Frequently Asked Questions

How do I set up OpenClaw on a Mac Mini?

Install Node.js v18+ via Homebrew, run 'npm install -g openclaw@latest', then run 'openclaw onboard --install-daemon'. Disable sleep in System Settings and enable auto-restart after power failure. Your Mac Mini will run OpenClaw 24/7.

What Mac Mini specs do I need for OpenClaw?

The base Mac Mini M2 or M4 with 8GB RAM is sufficient for OpenClaw. AI processing happens on cloud LLM servers, so you do not need a high-end configuration. The Mac Mini just runs the gateway, tools, and browser automation locally.

Can I run OpenClaw headless on a Mac Mini?

Yes, OpenClaw runs headless on a Mac Mini without a monitor. Access it remotely via SSH or screen sharing. The OpenClaw daemon runs in the background and your agents communicate through messaging channels like Telegram.

How do I keep OpenClaw running after Mac Mini restarts?

Install OpenClaw as a daemon using the onboarding wizard. This creates a launchd service that starts automatically on boot. Enable 'Restart automatically after power failure' in System Settings for maximum uptime.

Is a Mac Mini better than a VPS for OpenClaw?

A Mac Mini is better for long-term use because there are no monthly hosting fees after the initial $300 to $600 purchase. It also supports macOS-specific features like iMessage integration and Apple Reminders. A VPS is better if you want zero hardware maintenance.