OpenPaw

Your Own AI Assistant

A step-by-step guide to setting up OpenClaw — even if you've never opened Terminal
Time: about 30-60 minutes | Cost: ~$100-200/month | Difficulty: You can do this
Don't panic — you literally can't break your computer with these commands. Every step is explained in plain English.

What You're Building

By the end of this guide, you'll have a personal AI assistant that's more powerful than any subscription service — because it's yours.
🧠

An AI that knows your business inside and out

Train it on your processes, customers, and workflows. It remembers everything and gets smarter with every conversation.

📱

Text it from your phone (WhatsApp, Telegram)

Chat with your AI from anywhere — your car, a client meeting, or 11pm when you remember something important.

👁️

It can browse the web and research for you

Your AI has eyes on the internet. It can check websites, fill forms, and gather information automatically.

📊

A professional operations dashboard

Monitor your business, track metrics, and control everything from one clean interface you can access anywhere.

🔧

Automated tasks that run while you sleep

Set it once, let it work. Your AI handles routine tasks, sends reports, and keeps everything running 24/7.

💡

It gets smarter every day

Your AI builds a memory of your business — your preferences, processes, and decisions. The more you use it, the more context it has to work with.

What You Need

⚡ Before you start: Make sure Chrome is installed and your Mac Mini is plugged in and set to never sleep (we'll cover that in Step 6).
Item Cost Notes
Mac Mini (M1 or newer) $499-799 one-time Whisper quiet, runs 24/7, ~$3/month electricity
OpenAI / Codex account Plan varies The AI brain — connect with OpenAI/Codex OAuth so OpenClaw can use GPT-5.5
Internet connection (you have this) Needs to stay on
Google Chrome Free Required for AI browser control — download here
Domain name (optional) ~$12/year For professional URLs — requires a cloud server too
Cloud server (optional) $12-48/month For hosting dashboards online
Total minimum: Mac Mini + AI model access. Everything else is optional.

🤔 Mac Mini or Cloud Server?

Not sure which setup is right for you? We've got you covered.

View Side-by-Side Comparison →

See detailed pros, cons, costs, and step-by-step setup guides for both options

What is Terminal?

Don't Panic — Terminal is Just Texting Your Computer

Terminal is a text box where you type commands to your computer. That's it.

Think of it as texting your computer instead of clicking buttons. Instead of clicking "New Folder," you type mkdir new-folder. Same result, different method.

How to Open Terminal:

Mac: Press Cmd+Space, type "Terminal", press Enter
Windows: Search for "PowerShell", click it
Linux: You probably already know this 😄

Try These Right Now (They're Completely Safe):

pwd
"Where am I?" (shows your current folder)
ls
"What's in here?" (lists files)
clear
"Clean up this screen"

See? Nothing exploded. You're a Terminal user now.

Install Everything

Each command gets its own explanation. Copy, paste, and follow along — we'll walk through exactly what to expect.

Step 1: Install Homebrew (Mac only)

Homebrew is like an app store for Terminal. It helps install other tools.
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
What to expect: This takes 2-5 minutes. Follow any instructions it gives you at the end.
Password prompt: It might ask for your Mac password — that's normal. You won't see the characters as you type — also normal.

After it finishes, run the two commands it tells you (they look like):

echo 'eval "$(/opt/homebrew/bin/brew shellenv)"' >> ~/.zprofile
eval "$(/opt/homebrew/bin/brew shellenv)"

Step 2: Install OpenClaw

This one command installs everything OpenClaw needs:
curl -fsSL https://openclaw.ai/install.sh | bash
iwr -useb https://openclaw.ai/install.ps1 | iex
What to expect: Takes 1-2 minutes. It installs Node.js (the engine OpenClaw runs on) and OpenClaw itself.

Verify it worked:

openclaw --version

You should see a version number. If you do, you're golden.

Step 3: Run the Setup Wizard

openclaw onboard --install-daemon
This walks you through everything:
  • Choosing your AI provider (OpenAI/Codex recommended for GPT-5.5)
  • Setting up OAuth or an API key/token
  • Creating your workspace (where your AI stores its brain)
  • Optionally connecting WhatsApp, Telegram, or Discord
  • Installing auto-start (so it survives reboots)
Just follow the prompts — it explains each step.

Step 4: Connect OpenAI/Codex OAuth

OpenClaw can use OpenAI/Codex OAuth for GPT-5.5. Use this when you need to connect or recover the Codex login.

Run this to trigger the OpenAI/Codex device-code flow:

openclaw models auth login --provider openai-codex --method oauth
What happens: OpenClaw will show a device-code login flow. Complete it in the browser with the OpenAI account you want connected.

If OAuth does not launch:

openclaw models auth setup-token --provider openai-codex

Verify the connection:

openclaw models list
You want to see: openai-codex/gpt-5.5, auth oauth, and the expected OpenAI account, for example dr.gina78@gmail.com.

Step 4B: Give Every Model the Same Operating Standard

Smart models still need clear rules. When you switch from one model to another, paste this into your workspace so the new model acts like an operator: it reads context, uses tools, protects data, verifies work, and finishes the loop.

Create or edit your model handoff file:

nano ~/.openclaw/workspace/AGENTS.md

Paste this public-safe operating standard:

# Model-Agnostic Execution Standard

This applies to every model running in this workspace.

You are not just an advice bot. You are an execution-focused assistant.

## First principles
- Read the local context before acting: AGENTS.md, USER.md, STATUS.md, CONTINUATION.md, and any relevant project docs.
- Treat persistent user data as sacred: databases, uploads, credentials, .env files, documents, logs, and backups must not be overwritten during code changes.
- Use tools directly when a task is clear. Do not say you lack access until you have checked the available tools and context.
- Prefer evidence over guesses. Verify with tests, logs, curl, screenshots, database counts, or direct file inspection.

## How to act
1. Triage the request: routine, delegated, risky, or blocked.
2. For routine reversible work: do it, verify it, then report the result.
3. For destructive, external, costly, privacy-sensitive, or production-risk actions: stop and ask first.
4. Before changing a live system, make or confirm a backup. Never overwrite live databases, .env files, uploads, secrets, or user-generated records.
5. If changing code on a live service: edit the source, restart only the needed service, verify the live behavior, then back up/commit the changed files.
6. If the task grows complex, write a short handoff note before context gets heavy.
7. If blocked, state the exact blocker and the one missing thing needed to continue. Do not invent workarounds that risk data or trust.

## Communication style
- Be concise, specific, and useful.
- Do not bury the answer under a long plan if you can act.
- Report proof: command output, URL, commit hash, test result, screenshot, or named blocker.
- Say what changed, what was verified, and any caveat.

## Definition of done
A task is done only when:
- The requested outcome exists.
- The result has been verified.
- Persistent data is safe.
- Live changes are backed up and committed when applicable.
- The user has a short result summary with proof.

## Never do these automatically
- Delete data.
- Send external messages.
- Spend money.
- Publish broadly or deploy risky production changes.
- Disable security safeguards.
- Replace a database, .env file, upload directory, or credential store with a code artifact.

If instructions conflict, protect user data and ask for clarification.
Why this matters: A model switch should not change your operating standards. This gives GPT, Claude, or any future model the same practical rules: act when safe, ask when risky, protect data, verify results.

Step 5: Start Your AI

openclaw gateway start

Check that it's running:

openclaw gateway status
You should see 'running' — that means your AI is alive.

Step 6: Open Your Dashboard

openclaw dashboard

Or just go to: http://localhost:18789 in your browser

🎉 You're talking to your AI. Say hello.
Next step: Scroll down to download your operations dashboard — it walks you through setting up everything else (browser control, WhatsApp, knowledge base, and more).

Keep It Running 24/7

Your Mac Mini needs to stay on and awake for your AI to work.

Prevent Sleep

System Settings → Energy → "Prevent automatic sleeping" → ON

Or run:

sudo pmset -a sleep 0 displaysleep 5 disksleep 0 autopoweroff 0 standby 0
Password required: That's the sudo part.

Auto-Recover from Power Outages

System Settings → Energy → "Start up automatically after power failure" → ON

sudo pmset -a autorestart 1

Automatic Login

System Settings → Users & Groups → Automatic Login → Select your user

So it fully boots without waiting for a password

Verify Auto-Start

ls ~/Library/LaunchAgents/ai.openclaw.gateway.plist

If that file exists, OpenClaw starts automatically on boot.

If it's missing: openclaw gateway install

The Recovery Chain:
Power goes out comes back Mac turns on logs in OpenClaw starts AI reconnects no human needed

Optional: UPS Battery Backup (~$50-80)

Keeps everything running through brief power flickers

APC BE425M or CyberPower CP425SLG are good budget options

What's Next

Your AI is running. Now open your dashboard and work through the tabs:
🧠
System
You're here
💬
Chat
Talk to your AI
👁️
Browser
Give your AI eyes
👤
Personality
Teach it who you are
📱
Channels
Chat from your phone
🔑
Key Vault
Secure your keys
📚
Knowledge
Seed your AI's brain
💾
Config
Fine-tune behavior
📊
VPS
Go cloud (advanced)
🌐
Domain
Get a URL (advanced)
🚀
Deploy
Go live (advanced)

Troubleshooting

Problem Fix
Terminal says "command not found" Close Terminal, open a new one, try again
Homebrew install fails Run xcode-select --install first, then try again
openclaw not found after install Close Terminal, open new one. If still not found: npm install -g openclaw
Gateway won't start openclaw doctor to diagnose
Can't see the webchat Make sure gateway is running: openclaw gateway status
OpenAI/Codex OAuth expired or disconnected Run: openclaw models auth login --provider openai-codex --method oauth. Fallback: openclaw models auth setup-token --provider openai-codex. Verify with openclaw models list.
Mac Mini went to sleep Follow Section 5 to prevent sleep
Everything is broken Don't panic. openclaw gateway stop then openclaw gateway start

Starter Templates

Real full-stack apps built for a real business, genericized so you can make them yours. Preview them here, then download from your dashboard.
🚨

Triage Command Center

A React + Node.js Kanban board for managing appointments and tasks across multiple days. Color-coded priority flags, task checklists, team assignment, shift management, SLA tracking, and optional AI-powered triage.

Stack: React 18, Vite, Tailwind CSS, Express, SQLite

👀 Preview
📖

Team Handbook

A searchable, AI-powered team handbook. Rich text editor, hierarchical sections, file uploads, version history, and role-based access. Your team gets instant answers with AI search instead of digging through binders.

Stack: Node.js, Express, SQLite, Quill Editor, OpenRouter AI

👀 Preview
🧠

Teaching Your AI to Remember

How memory works, what files to create, and copy-paste prompts to train your AI — without burning through usage limits.

Includes: 5 setup prompts, the 3-tier memory system, monthly health check template

📖 Read the Guide →

Get Your Dashboard

Your operations dashboard — the same design used by a real veterinary practice. It walks you through setting up everything else.

👀 Preview It ⬇️ Download Mac App ⬇️ Download Windows
After downloading:
  1. Double-click the file to open it in your browser
  2. Work through each tab to set up your AI
  3. Or tell your AI: "Help me customize my dashboard for my business"