Five steps.
Two prompts, one detour to your phone. Steps 1–4 get it alive. Step 5 is the one that makes it yours.
Step 1
Get ChatGPT.
The only part that costs money, and there's a good chance you already pay for it. One subscription, one login, no bill to watch.
ChatGPT Plus
$20/mo · chatgpt.com/download
- Most people already have it
- Signs in with the account you already use — never an API bill
- Its coding side, Codex, is what builds this for you in step 3
- Plenty for chatting on demand. Upgrade later only if you start running things around the clock
Why only one option? Other AI subscriptions work too, and you can add one later in about a minute by asking your assistant. But every extra choice on this page is another chance to get stuck on step one. ChatGPT is the lane that's officially supported end to end, and it's the one most people are already paying for.
Step 2
Give it a Mac that stays awake.
This is the one real requirement, and it's worth getting right. Your assistant is only alive while its computer is. A sleeping Mac is an assistant that isn't there.
You don't need a new machine or a spare one. You need one machine that's plugged in and not napping. Find yourself below:
Desktop Mac
Mac mini, Studio, iMac. Easiest case by far — one setting and you're done. This is what most people end up with eventually.
Laptop, lid open
Plugged in, lid up, sleep set to never. Totally fine, and how nearly everyone starts. It just has to live somewhere it can stay open.
Laptop, lid closed
Close your laptop at night or while you’re away? Keep it plugged in and it can keep working with the lid closed. You do not need a monitor; step 3 walks you through the one free app that makes it work.
About the closed-lid one. Keep it plugged in. macOS normally refuses to stay awake with the lid shut unless a display and a keyboard are attached. A free Mac App Store app called Amphetamine removes that requirement, and turns the internal screen off so it isn't burning in behind the lid. On Apple Silicon MacBooks turn on its Power Protect option too, or plugging and unplugging the charger can drop the session. Step 3 walks you through it.
If it has to sleep at night, that's still fine. You get a fully working assistant that's simply offline while you are. The only thing you lose is work happening overnight — briefings, scans, anything scheduled. That's the honest reason to add a cheap always-on Mac later: because you got hooked, not because you're doing it wrong now.
Two things to know. A laptop working hard with the lid shut runs hotter, so give it air and don't bury it in a drawer. And on Windows this all works too — power settings are under Settings → System → Power, and the prompt handles it.
Step 3
Paste this. Watch it build itself.
In the top-left menu of the desktop app from step 1, choose Codex. Start a new task, let it run things on your computer (that's the next section), then paste this in and send. It installs everything, keeps your Mac awake, wires up your phone, and checks its own work before it tells you it's done.
This one gets it alive. It deliberately doesn't touch your email or messages yet — that's step 5, and it goes much better once you've already seen the thing reply to you.
I want to set up OpenClaw on this computer so I have my own always-on AI
assistant that I can text from my phone. Please do the entire setup for me.
Assume I am not a developer. Do not ask me to make technical decisions, and
explain each step in plain language as you go. Do the work yourself rather
than handing me instructions to follow.
1. Read the official guide at https://docs.openclaw.ai/start/getting-started
first and follow that, rather than anything you remember. It changes.
2. Install it by running:
curl -fsSL https://openclaw.ai/install.sh | bash
That script brings its own Node and anything else it needs. Do not install
Homebrew, Node, Git, a code editor or a terminal app separately.
3. The installer opens a setup wizard on its own. Walk me through it. When it
asks about the AI provider, choose OpenAI and sign in with my existing
ChatGPT SUBSCRIPTION. Do NOT set up a pay-per-use API key, and do not
create any billing account.
4. Make it run in the background and survive a restart:
openclaw gateway install
5. Keep this machine awake, because the assistant only runs while it is.
Work out what kind of computer this is and set it up to match:
- A desktop Mac: set the computer to never sleep while on power.
- A laptop I keep open: never sleep while plugged in, and tell me to leave
it plugged in with the lid up.
- A laptop I close at night or when I'm away, with the lid CLOSED: tell me
to keep it plugged in and install the free app Amphetamine from the Mac App Store,
then walk me through turning off
"Allow system sleep when display is closed" and turning on Power Protect
if this is an Apple Silicon Mac. Warn me it runs hotter closed, so it
needs air.
Ask me which of these I want rather than guessing. Then tell me which case
I am, show me the setting you changed, and tell me how to undo it.
6. Set up Telegram so I can text it from my phone. Read
https://docs.openclaw.ai/channels/telegram and follow it. Walk me through
creating a bot with @BotFather on my phone one message at a time, wait for
me to paste you the token, then put the token into the OpenClaw config
yourself and restart the gateway. Set the DM policy to pairing so strangers
cannot talk to it. When I send my first message, run
"openclaw pairing list telegram" and approve my code with
"openclaw pairing approve telegram <CODE>".
7. Prove it works. Run: openclaw gateway status
Then have me send a real Telegram message and confirm I get a reply back.
Do not tell me it is done until you have seen it reply. If something
breaks, run: openclaw triage
Rules: stop and ask me before anything that costs money, sends a message to
another person, or deletes files. At the end, give me a short list of what
you set up, where it lives, and three things I should try first. Don't connect
it to my email, calendar or messages yet — that's the next prompt.
Step 4
Pick up your phone.
This is the part your computer can't do for you, and it's the part that makes the whole thing feel real. You're giving your assistant a phone number, in the form of a Telegram bot that only answers to you.
Step 3 will walk you through this in real time. Here's the whole thing anyway, so nothing surprises you:
1. Get Telegram, if you don't have it
Free, on the App Store. Sign up with your phone number. You'll never need to use it for anything else.
2. Message @BotFather
Search Telegram for BotFather — the one with the blue checkmark — and open the chat. This is Telegram's official robot for making robots. Tap Start.
3. Send /newbot
It asks for a display name — anything you like, this is what shows at the top of the chat. Then it asks for a username, which has to be unique and end in bot. Expect to be told your first few are taken.
4. Copy the token it gives you
A long line that looks like 7891234567:AAF.... Paste it back to the assistant on your computer and it does the rest — config, restart, all of it.
5. Text your bot
Open the chat with your new bot and say hi. Nothing comes back yet — that's on purpose. It's waiting to be told you're you.
6. Approve yourself, once
Tell the assistant on your computer that you sent it. It approves your account, and from then on the bot replies to you and ignores everyone else. That's the last step. Say something to it.
/revoke in BotFather issues a fresh one.
Why Telegram first, and not iMessage? It's the only one that hands you a bot in two minutes with no business account, no approval queue, and no permissions to grant. iMessage genuinely works too, and step 5 sets it up if you want it — but it needs macOS permissions that are a bad first experience. Get one channel working, then add the other.
Step 5
Now connect it to your life.
Everything so far gave you a very capable stranger. This is the step that turns it into your assistant — your mail, your calendar, your messages, your notes, and a memory of who you are.
Here's the trick most people miss: you don't have to know what to ask for. This prompt interviews you first. It asks what you use and what you want, one question at a time, then goes and wires up only the things you said yes to. You're not expected to know the options — that's its job.
You're running now. Next I want to connect you to the things I actually use, so you're useful to ME instead of being a generic assistant. Interview me FIRST. Ask me the questions below one at a time, in plain language, and wait for my answer before moving to the next one. Don't assume anything about my setup — if you don't know, ask. I am not technical, so ask me about my life, not about software. Ask me: 1. Who I am, what I do for work, and the people who matter most day to day. 2. What I'd love your help with in the first week. 3. What email I use — Gmail, iCloud, Outlook, a work account, or several — and whether I want you reading it. 4. Whether my calendar and contacts live in Google or Apple. 5. Whether I want you to read and send iMessages from this Mac. 6. Whether I use Apple Notes, Apple Reminders, or something else for to-dos. 7. Where my important files and photos actually live. 8. Anything you should never touch, and anything you must always ask me about before doing. Then set up ONLY the things I said yes to. Run "openclaw skills check" to see what's already available to you, and read https://docs.openclaw.ai for the real instructions rather than trusting your memory. A few that people get wrong: - Google mail, calendar or contacts: use the Google Workspace skill and sign me in through the browser. Never ask me to create an API key or a billing account. - iCloud or other IMAP mail: use the mail skill. It needs an app-specific password from account.apple.com, NOT my real Apple password. Walk me through creating one. - iMessage: install imsg and enable OpenClaw's iMessage channel. I'll have to grant Full Disk Access and Automation in System Settings — tell me exactly where to click. Basic send and receive only. Do NOT disable System Integrity Protection on my Mac for the advanced features, and don't ask me to. - Apple Notes and Apple Reminders each have their own skill. Add them if I said I use them. Rules while you do this, no exceptions: - Set everything up READ-ONLY first. Prove you can read my mail and calendar before you have any ability to send or change anything. - Never send a message, reply to an email, or accept an invitation as me without asking me first — every single time, until I tell you to stop. - Every password, token and key goes in a password manager or the system keychain. Never in a chat, a note, or a file in a repository. - If something needs a permission only I can grant, stop and give me click-by-click directions instead of trying to work around it. Finally: write everything you learned about me into your own memory files so you never have to ask again. Test each connection with something real and show me the result. Tell me plainly which ones did NOT work and why. Then give me five things to try that you couldn't have done an hour ago.
Expect a couple of "go click this" moments. Reading your Messages or your calendar needs a permission macOS only lets you grant, in System Settings. Your assistant will tell you where to click, but it genuinely cannot do those for you. That's the operating system protecting you, and it's working as intended.
Say no to plenty. There's no prize for connecting everything on day one. Pick the two things you'd actually use this week, and add the rest whenever they come up — it's the same conversation every time.
