If you just bought a Mac Mini and want a reliable AI assistant running 24/7, this is the setup.
This guide is built from real mistakes, real fixes, and real operating experience. It is written for people who do not want to babysit a terminal.
You will get:
- A Mac Mini that stays awake and recovers from outages
- OpenClaw running as always-on infrastructure
- Secure remote access through VPN
- Safe credential handling with 1Password
- Lower monthly spend with the right model defaults
- Practical prompts you can copy and use immediately
- A troubleshooting section for common setup failures
Before You Start
You need:
- Mac Mini on power (Ethernet preferred)
- Admin access on the Mac
- ChatGPT or Claude subscription (not required, but will save you a ton in API credits)
- Tailscale account
- 1Password account
Step 1: Stop the Mac Mini from Sleeping
If your Mac sleeps, OpenClaw is offline.
1.0 Open Terminal first
- Press
Command + Space - Type
Terminal - Press Enter
You should see a blinking cursor. That is where you type commands.
1.1 Click settings first
Open System Settings and set:
- Battery / Energy: Prevent automatic sleeping on power adapter = ON
- Battery / Energy: Wake for network access = ON
- General / Login Items: confirm nothing is forcing sleep
- Lock Screen: Turn display off on power adapter = Never
1.2 We will run one terminal command in Step 2
You just finished the click-settings part.
In Step 2 (right after opening Terminal), we will paste one command to lock these power settings in place.
Step 2: Install OpenClaw (5th-Grade Version)
2.1 Open Terminal
- Press
Command + Space - Type
Terminal - Press Enter
You should see a blinking cursor. That is where you type commands.
2.1A Run the no-sleep command (from Step 1)
Paste this now:
sudo pmset -a sleep 0 disksleep 0 displaysleep 0 powernap 0 autorestart 1 womp 1
pmset -g
What this gives you:
- No system sleep
- Auto restart after power failure
- Wake-on-LAN support
Think of this like setting up a game console.
We are installing 3 things in order:
- Homebrew (an app store for terminal tools)
- Node.js (the engine OpenClaw needs)
- OpenClaw itself
2.2 Install Homebrew (skip if you already have it)
Paste this and press Enter:
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
If Mac asks for your password, type it and press Enter. (You will not see letters while typing. That is normal.)
Now test Homebrew:
brew --version
If you see a version number, you are good.
2.3 Install Node.js
brew install node
Check it worked:
node -v
npm -v
2.4 Install OpenClaw
curl -fsSL https://openclaw.ai/install.sh | bash
openclaw --version
If you see a version number, OpenClaw is installed.
2.5 Run setup wizard
openclaw onboard --install-daemon
The wizard will guide you through login and setup.
2.6 Confirm it is running
openclaw gateway status
openclaw status
If you see running/healthy status, this step is done.
Common install errors (quick fixes)
-
brew: command not found- Homebrew did not install correctly. Re-run the Homebrew install command.
-
openclaw: command not found- Close Terminal, reopen, run
openclaw --versionagain. - If still missing, run the OpenClaw install command one more time.
- Close Terminal, reopen, run
-
Password input looks frozen
- Terminal hides password characters by design. Type carefully and press Enter.
Step 3: Connect Your Existing ChatGPT or Claude Subscription First
Before you think about API credits, do this.
For most people, the easiest and cheapest way to start is connecting an existing subscription during onboarding.
What to choose in onboarding
When openclaw onboard --install-daemon asks about auth, choose one of these:
- OpenAI / ChatGPT route if you already use ChatGPT
- Anthropic / Claude route if you already use Claude
Then complete the browser login prompt.
Verify your connection
After onboarding, run:
openclaw status
You should see your provider authenticated and the gateway running.
Important note
Do not worry about detailed model defaults yet.
First get OpenClaw installed and working end-to-end.
We will set model routing and cost policy later, after everything is stable.
Step 4: Set Up Secure Remote Access with Tailscale
Do not expose your Mac Mini directly on the public internet.
Install and start Tailscale:
brew install --cask tailscale
brew services start tailscale
Authenticate and confirm connection:
tailscale status
tailscale ip -4
Use your Tailscale hostname/IP to reach the machine remotely.
Step 5: Set Up 1Password for Secrets
Use 1Password for every credential. Do not keep secrets in random text files.
Store:
- API keys
- OAuth tokens
- WordPress logins
- Domain registrar logins
- Service credentials (n8n, Slack, Google, etc.)
Naming convention that stays clean:
Service - Environment - Purpose
Examples:
OpenAI - Prod - Codex AuthWordPress - losalacranesranch.com - AdminN8N - Cloud - API Key
Step 6: Storage Optimization Rules
Most people ignore this until things break.
Set these rules now:
- Keep at least 20-30 GB free at all times
- Remove duplicate media/backups quickly
- Move giant archives off primary disk when done
Quick checks:
df -h /
du -sh ~/Downloads/* 2>/dev/null | sort -h | tail -20
If a file is done and recoverable elsewhere, remove it.
Step 7: Build a Useful Workspace Structure
In your OpenClaw workspace, keep:
SOUL.mdfor assistant behavior and toneUSER.mdfor your context and goalsMEMORY.mdfor long-term factsmemory/YYYY-MM-DD.mdfor daily logstodos/life/active-todos.jsonfor personal taskstodos/business/active-todos.jsonfor work tasks
This creates continuity after compaction and session resets.
Step 8: Memory Hardening (Critical Fixes from Real Use)
This is the part most setups miss.
Without memory hardening, your assistant will forget tasks after compaction, lose priorities, and drift over time.
Use this file system:
MEMORY.md= curated long-term memorymemory/YYYY-MM-DD.md= daily log and decisionstodos/life/active-todos.json= life taskstodos/appsumo/active-todos.json= work taskstodos/business/active-todos.json= business tasks
Hard rules:
- If it should be remembered, write it to a file immediately
- Never trust "mental notes" inside one session
- Store todos in JSON files, not chat history
- Check memory + todos at the start of each session
- Keep memory updates short, factual, and timestamped
Copy this prompt to enforce it:
From now on, if I say "remember this" or assign a task, write it immediately to the right memory or todo file.
Do not confirm until the file is actually updated.
At the start of each session, read MEMORY.md and active todo files before planning.
Why this matters:
- Survives context compaction
- Survives model switches and restarts
- Prevents dropped tasks and repeated conversations
Step 9: Real Use Cases You Can Run Today
Prompt 1: Daily operator briefing
Give me my daily operator briefing:
- Top 3 priorities from my todo files
- Deadlines in next 7 days
- Any blocked items needing a decision
- One highest leverage action for today
Keep it concise.
Prompt 2: Build reminders from raw messages
Convert this message thread into actionable todos.
For each item include: title, priority, deadline, and next action.
Then update the correct todo file (life, business, or appsumo).
Prompt 3: Cost check and model audit
Audit my current model usage and credit risk.
Tell me where expensive models are being used unnecessarily.
Propose exact downgrades to Codex/Sonnet.
Prompt 4: Weekly systems health review
Run a weekly systems check:
1) openclaw status
2) gateway health
3) storage pressure
4) failing automations
5) stale credentials
Return fixes in priority order.
Prompt 5: Personal assistant mode
Review my life todos and memory files.
Create a short plan for this week with:
- What to do first
- What can wait
- What should be delegated
Step 10: Set Model Defaults (After Setup Is Working)
Now that setup is stable, set model policy.
Use this rule:
- Use your connected ChatGPT or Claude subscription path first
- Keep your day-to-day model on the lower-cost option
- Only use high-end models for hard tasks
Copy this to your assistant:
Set a cost-safe model policy for me.
Use my connected subscription path by default.
Use cheaper models for normal tasks.
Only use high-cost models when complexity truly requires it.
Keep this as a persistent rule.
Check it took effect with:
/status
Step 11: Backup and Recovery Habits
Every week:
- Confirm OpenClaw gateway is healthy
- Confirm VPN is connected
- Confirm important credentials are in 1Password
- Confirm disk has safe free space
- Confirm your todo files are still current
If something breaks, restart in this order:
openclaw gateway restart
openclaw status
Then check model auth and channel auth.
Step 12: Common Errors and Fixes
Error: assistant stopped responding overnight
Likely cause: Mac slept or rebooted without service recovery.
Fix:
- Re-check
pmsetsettings - Confirm auto restart settings
- Restart gateway:
openclaw gateway restart
Error: credits burning too fast
Likely cause: expensive model running by default.
Fix:
- Set Codex as default
- Use Sonnet as secondary
- Use Opus only when explicitly needed
- Re-check model policy weekly
Error: browser automation says no tab connected
Likely cause: browser relay tab not attached.
Fix:
- Open Chrome tab
- Click OpenClaw Browser Relay extension icon
- Ensure badge shows connected
- Retry action
Error: secrets all over the place
Likely cause: no secret discipline.
Fix:
- Move all credentials into 1Password
- Remove plaintext duplicates
- Use consistent item naming
Error: disk suddenly full
Likely cause: backup zips and media duplicates.
Fix:
- Find largest folders/files
- Delete or archive completed backups
- Keep free space above 20 GB
What This Setup Usually Costs
Rough monthly ranges:
- OpenClaw model usage: highly variable, reduced significantly with Codex-first policy
- Tailscale: free tier is enough for many solo setups
- 1Password: standard account pricing
- Hosting hardware: Mac Mini already owned
The biggest variable is model selection discipline.
Final Notes
If you do only five things, do these:
- Disable sleep completely
- Keep Codex as default
- Keep Opus on a short leash
- Use Tailscale instead of public exposure
- Store every secret in 1Password
That is the difference between a toy setup and reliable personal infrastructure.
If you want, I can publish a part two with a full "OpenClaw for business operations" playbook, including:
- lead follow-up automations
- daily KPI anomaly detection
- social content pipelines
- executive briefings and task management