Bot Setup Guide

Build a personal AI that runs 24/7 on your own machine. Each phase unlocks a new capability. Check off steps as you go — progress saves automatically.

Setup

Get a dedicated machine running with OpenClaw and remote access. After this, the AI is alive but you haven't talked to it yet.

Get a dedicated computer

Your AI needs its own machine that never turns off. The best option is a Mac mini — it's small (about the size of a sandwich), silent, and uses very little electricity ($2-3/month). A spare laptop with the lid closed also works. If you don't have spare hardware, you can rent a cloud server (VPS) from Hetzner or DigitalOcean for $5-10/month.

This should not be the computer you use every day. Your AI gets its own machine so it can run 24/7 without interrupting your work.

Install Homebrew

First, open Terminal. Terminal is an app on your Mac that lets you type commands directly to the computer. To open it: press the Command key (⌘) and the Space bar at the same time. A search bar called Spotlight appears. Type terminal and press Enter. A window with a blinking cursor opens — this is Terminal.

Copy the line below (click to copy), go to Terminal, paste with Command + V, and press Enter.

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

It asks for your Mac password. When you type it, nothing appears on screen — no dots, no stars. That's normal. Type it and press Enter. Wait 2-5 minutes until you see a new blinking cursor.

Prevent the computer from sleeping

Your AI can't work if the machine sleeps. Open Terminal and paste:

copysudo pmset -a sleep 0

sudo means "do this with administrator permission." Password prompt works the same (invisible typing).

Also: System Settings → Displays → Advanced → turn off "Automatically adjust brightness." For laptops: System Settings → Battery → Options → "Prevent automatic sleeping on power adapter."

Create a Gmail account for the AI

Your AI needs its own email to sign up for services. Go to Gmail signup on your phone (easier for Google's phone verification). Pick any name — it's a throwaway identity.

Then on the AI's computer, open a browser, go to gmail.com, log in with the new account, and leave it logged in.

Never use your personal Gmail. The AI's accounts are completely separate. If something goes wrong, only a throwaway email is at risk.

Install OpenClaw

Open Terminal and paste:

copycurl -fsSL https://openclaw.ai/install.sh | bash

This installs OpenClaw, registers it as a background service (auto-starts on boot), and starts it immediately. It keeps running even if you close Terminal or restart.

Alternative if you have Homebrew: brew install openclaw then openclaw gateway install && openclaw gateway start.

Install Tailscale for remote access

Tailscale (free) creates a private encrypted network between your devices, no matter where they are. You'll use it to access the AI's screen and dashboard from your phone or laptop — even from another country.

copybrew install --cask tailscale

Open Tailscale from Applications. Click the menu bar icon (top right) and sign in with Google, Microsoft, or GitHub. The machine gets a Tailscale IP (like 100.x.x.x) — write it down.

Then install Tailscale on your phone (App Store / Play Store) and laptop (tailscale.com/download). Sign in with the same account on each.

Enable screen sharing now. On the AI's computer: System Settings → General → Sharing → turn on Screen Sharing. From your Mac: Finder → Go → Connect to Server → vnc://[Tailscale IP]. From your phone: use a VNC app like Screens (iOS) or bVNC (Android). You'll need this when the AI hits CAPTCHAs or browser issues.

Enable the OpenClaw dashboard

OpenClaw has a web dashboard for checking status, viewing conversations, and managing settings.

copyopenclaw config set gateway.controlUI.enabled true openclaw gateway restart

From any device on your Tailscale network, open a browser and go to http://[Tailscale IP]:18789. Bookmark it on your phone for quick health checks.

The machine is ready. OpenClaw is running, Tailscale gives you remote access, and the dashboard lets you monitor everything. Next: connect a messaging app so you can talk to it.

Chat

Connect a messaging app and start talking to your AI. After this, you have a working assistant.

Connect Telegram

Telegram is a free messaging app. This is how you'll talk to your AI — you send a message, it replies in the same chat, like texting a friend.

Install it on your phone from the App Store or Play Store. Then follow the OpenClaw Telegram setup guide (~5 minutes — you create a "bot" in Telegram and paste a token into OpenClaw).

Other options: OpenClaw also works with Signal, Discord, Slack, WhatsApp, and iMessage. Telegram is the easiest to start with.

Send your first message

Open Telegram and message your AI: "Hello, are you there?"

It should reply within seconds. If not, check the dashboard (Setup phase) for errors. Common fix: openclaw gateway restart in Terminal.

Give it a real task right away. Don't spend a week configuring. Ask something real: "Build me a personal website." "Research hotels in Kyoto for April." "Summarize this PDF." You learn by using it, not by tweaking settings.

Let the AI create its own accounts

Tell the AI to sign up for developer services. Send it:

"Sign up for GitHub, Vercel, and Supabase using your Gmail account."

It opens a browser, goes to each site, and handles registration. Help with CAPTCHAs via remote screen access from Setup.

For small purchases (like a $11/year domain), create a free virtual card at Privacy.com with a $20/month limit. You type the card number yourself.

Share your existing projects

Invite the AI to collaborate:

GitHub: Repository → Settings → Collaborators → Add the AI's username.

Vercel: Team → Members → Invite the AI's Gmail.

Supabase: Organization → Members → Invite the AI's Gmail.

Google Docs: Share → Enter the AI's Gmail.

You now have a fully working AI assistant. It can write code, deploy websites, manage databases, search the web, and research anything. Everything after this makes the experience better — none of it is required.

Sync

See everything your AI writes on all your devices. Encrypted, beautiful, searchable by meaning.

Install Obsidian on the AI's computer

Obsidian is a free app that makes Markdown files (your AI's native format) look beautiful. Open Terminal and paste:

copybrew install --cask obsidian

Open Obsidian (Command + Space, type "obsidian"). Click "Open folder as vault." Press Command + Shift + G, type ~/.openclaw/workspace, click Open. You'll see all the AI's files.

Enable Obsidian Sync

Obsidian Sync ($4/month) encrypts your files end-to-end and syncs across all devices. Not even Obsidian can read your content.

In Obsidian: gear icon (⚙️) → Core plugins → toggle on Sync → click Sync in sidebar → log in → Create new remote vault.

In Sync settings, find "Excluded folders" and add Code. Code repos use Git, not Sync.

Leave "Sync all file types" off. Default (Markdown only) is correct. Code projects have thousands of build files that would overwhelm storage.

Keep Obsidian running forever

Obsidian only syncs while open. Auto-launch and self-restart it:

copycat > ~/Library/LaunchAgents/md.obsidian.keepalive.plist << 'PLIST' Labelmd.obsidian.keepalive ProgramArguments open-aObsidian RunAtLoad StartInterval300 PLIST launchctl load ~/Library/LaunchAgents/md.obsidian.keepalive.plist

Launches on boot, restarts every 5 minutes if closed.

Connect your phone and laptop

Phone: App Store or Play Store. Laptop: obsidian.md/download.

On each device: new vault → Settings → Sync → same account → select the remote vault. Everything the AI writes appears on your phone within seconds.

Create a Whiteboard.md at the vault root. "W" sorts to the bottom, always easy to find. Shared scratchpad: paste links, jot thoughts, dump ideas. Say "check my whiteboard" and the AI knows what you mean.

Enable semantic search

Lets your AI search all notes by meaning, not keywords. "What restaurant did James recommend?" works even if "restaurant" never appears. Runs locally — no data sent anywhere.

copyopenclaw config set agents.defaults.memorySearch.provider local openclaw config set agents.defaults.memorySearch.fallback none

Tell it which folders to search. Open the config:

copyopen ~/.openclaw/openclaw.json

Find memorySearch and add your folders:

"extraPaths": ["Contacts", "Lists", "Memos", "Tasks"]

Save, close, restart:

copyopenclaw gateway restart

First run downloads a model (~313MB) and indexes notes (2-5 min). After that, instant.

Keep folders one level deep. Contacts, Lists, Memos, Tasks — every note directly in one folder. If you can't find it by filename, rename it. Tell the AI your system and it files things correctly.

Put all code in Code/. Every repo, project, and skill. Excluded from Obsidian Sync — code uses Git, notes use Sync. Tell the AI: "Always clone into Code/."

Let the AI finish its own setup. Paste this page's URL into Telegram and say "follow this guide." It can complete the remaining steps itself.

Harden

You've given an AI real tools. Lock it down so the risk stays manageable.

Dedicated machine, separate accounts

The AI should never run on your personal computer or be logged into your personal accounts. Its own machine, Gmail, GitHub, everything. If compromised, the damage is a throwaway Gmail and free-tier accounts — not your personal life.

Gateway stays local

The OpenClaw gateway only listens on 127.0.0.1 (localhost). No one on the internet can reach it. Never expose it publicly. Tailscale handles remote access through a private tunnel.

Write safety rules

Open SOUL.md in your AI's workspace (visible in Obsidian) and add rules:

"Never send emails, make purchases, or post publicly without asking me first. Never access files outside the workspace. Use trash instead of permanent delete."

Like setting expectations with a new employee. Not unbreakable, but significantly reduces unintended actions.

Use scoped tokens with expiration dates

API tokens should have minimum permissions and expire after 90 days. Ask your AI: "Review your API tokens and flag any with excessive permissions or no expiration."

Keep OpenClaw updated

New features and security patches ship frequently. Run weekly:

copyopenclaw update

Or tell the AI: "Run openclaw update."

What's still risky: Prompt injection (malicious content tricking the AI) is unsolved industry-wide — that's why throwaway accounts matter. The LLM provider sees your conversations, so don't share passwords or financial data. Only install OpenClaw skills from trusted sources.