Which of these can you start from a phone with no computer open: OpenClaw, Grok Bot, or a Dot?

Grok Bot is the one you can start from a phone with no computer open. xAI’s mobile docs say a new user opens the iPhone or Android app, signs in, picks a first Bot, and waits while the shared cloud computer is set up. A Dot cannot be created on mobile: OpenAI’s getting-started article says you create it in the ChatGPT desktop app or in ChatGPT on desktop web, and that you cannot currently create a Dot on mobile or on mobile web. OpenClaw’s phone apps do not host the Gateway. With every computer off and no Gateway already running somewhere else, the phone has nothing to control.

The three products all have a phone surface. Only one of them treats the phone as a place where the first setup happens. The other two use the phone as a remote for work that was started on a computer, or that is still running on a computer you cannot see.

What does the Grok Bot phone app actually start?

It starts a Bot on the shared cloud computer, not a process on your laptop. The mobile page says the app connects to the same Bots, conversations, routines, connectors, and shared cloud computer as the desktop app, and that work continues in the cloud when the app is closed. Requirements are a phone (iOS 18 or later, or Android 9 or later; the iOS app also runs on iPad), an eligible plan, and an internet connection. The plan list is the same one the rest of the Grok Bot docs use: a paid individual Cursor plan, Cursor Teams, or a linked SuperGrok, SuperGrok Plus, or SuperGrok Heavy subscription. There is no line that says a Mac or a PC has to be awake.

The sign-in steps are on the phone. Open Grok Bot, choose Log In or Sign Up, finish Cursor authentication in the browser, and come back. If access comes from SuperGrok, the access screen offers Link Grok Account, then Finished Linking? Refresh My Status. A new user can finish the first-run tour, choose a first Bot, and wait for the shared computer. An existing user lands on the synced Bot list. The + control on the home screen creates a New Bot or a New Group Chat. From a conversation you can send text, use Start dictation, use Start voice chat, attach a photo or a file, and, on iPhone, send something in through the system share sheet.

That is a real start. You are not opening a chat window that only drafts. The computer the Bot works on is in Cursor’s cloud, and every Bot on the account shares it. Watching that screen, taking over for a password, a two-factor code, or a CAPTCHA, and handing control back are all listed on the mobile page. The screen you open from the phone is that shared computer.

A closed laptop still matters if you wanted the Bot on your own machine. Local computer execution is a separate approval from the cloud computer. I have not read a mobile page that says the phone wakes a shut laptop and runs commands there. If the only computer in the house is asleep, cloud work can proceed. Work that needed that laptop’s files and shell cannot. Some controls stay on the desktop even when the cloud job is fine: editing a routine’s schedule or instruction, and testing a routine, currently require the desktop app. Teach-by-demonstration is in the same bucket. The mobile page says some advanced desktop controls are not available on the phone. Starting a Bot and handing it a task is not one of those missing controls.

Why a Dot still needs a desktop the first time

Because creation is desktop-only, even though later messages are not. The getting-started article says to create the Dot in the ChatGPT desktop app or in ChatGPT on desktop web, follow the onboarding prompts, and that the desktop app is also available on Windows. After that, you can talk to it in the ChatGPT mobile app when mobile access is available. The same article says you cannot currently create a Dot on mobile, and that Dots are not supported on mobile web. The 29 September 2026 announcement says the same shape in shorter form: create the first Dot in the desktop app or the desktop browser, and message it in the mobile app after the initial setup.

So “no computer open” fails for the first Dot. A phone with ChatGPT installed, and no desktop session anywhere, cannot complete the step the help article requires. Once the Dot exists, the phone is a way to talk to a cloud computer that already has files, software, and browser sessions. Cloud work can continue while your devices are off. That is a different question from starting. Closing ChatGPT does not answer whether you were allowed to create the Dot from the couch.

Phone messaging after setup has extra conditions I should not flatten. Mobile access is “when available.” Texting, if it is available for the account, is connected from the desktop app; the help text says you use the desktop app to connect the phone and follow the prompts, and only then message the Dot from the phone. Enterprise phone messaging over iMessage, RCS, or WhatsApp was not available at launch in the workspace notes I read. If your account does not show mobile or texting, the desktop remains the only door. I am not going to invent a mobile create button that the help article says does not exist.

The Dot’s own computer is also not your laptop. Connecting your computer is optional, starts off, and is a desktop-app step. While the Dot uses that machine, the machine has to stay online with the ChatGPT app open. A phone in your pocket does not substitute for that. Revoke access is what removes the local path. Leaving the laptop shut simply means the Dot cannot use it. The cloud computer can still run.

What has to be running before an OpenClaw phone is useful?

A Gateway on a computer. The Android docs state the role in one line: companion node, and Android does not host the Gateway. The Gateway has to run on macOS, Linux, or Windows via WSL2. The iOS app’s prerequisite is a Gateway running on another device, reached over the same LAN, a tailnet, or a manual host and port. Nodes do not run the Gateway service. Channel messages land on the Gateway, not on the phone.

That is why a phone with no computer open cannot start OpenClaw. There is no cloud computer in this product for the phone to wake. The model call, the channels, and the session state live where the Gateway lives. If that host is off, the Telegram chat, the WhatsApp plugin, and the iOS app are all talking to a socket that is down. The phone app controls a computer somewhere else. It does not become that computer because the laptop lid is shut.

The exception is a host you already left running. A home server or a VPS with the Gateway installed as a service can stay up while every laptop in the house is closed. In that case the phone can message an agent that is already on. You did not start OpenClaw from the phone. You reached a process that was started on the host, with openclaw gateway install or an equivalent service, before you sat down. Pairing still happens against that Gateway. The Android and iOS docs both expect a running Gateway and a path to it before the phone connects. A first-time install, the config file, and the model credentials are host work. Doing them “from the phone” is not a documented path.

Sleep is not the same as a service on another box. If the only Gateway is on the laptop you just closed, and that laptop sleeps, the agent stops with the machine. The phone cannot unsleep it. Tailscale, SSH, and a LAN address only help when the Gateway process is listening.

Which jobs still fail on the one product you can start?

Anything that is not on the shared cloud computer. Grok Bot from the phone can begin a cloud task, approve a step, and review a result. It cannot, from the pages I read, edit and test a routine, or record a teach-by-demonstration session. Those stay on the desktop. A password, a passkey, or a payment still belongs in a takeover on the computer screen, not in an ordinary chat message. The mobile page tells you to open the computer and take control for a password, a two-factor code, or a CAPTCHA. That is a tap on the phone that opens the cloud screen. It is not permission for the Bot to read a secret you typed into the thread.

Plugins and connectors follow the account. Installing or reviewing plugins is listed under mobile Settings. A connector you add is on the shared computer the other Bots already use. Starting from the phone does not create a private computer for that one Bot. If the job needs a login you do not want every Bot on the account to be able to reach, the phone is the wrong place to “just sign in once” and walk away.

Push notifications are not the start button either. The mobile docs say to enable them if you want to hear about a result, a question, or an approval, and that push delivery is still rolling out. In-app attention states remain when push is not enabled. A Bot can be working while the phone is locked. The notification is how you hear about it. It is not what keeps the cloud computer alive.

So which one do you open when the laptop stays shut?

Open Grok Bot if the work can live on the shared cloud computer and you are willing to use the desktop later for routine edits and for teaching by showing. Do not open ChatGPT on the phone expecting to create a Dot. Create the Dot on a desktop first, then use the phone to message the one that already exists. Do not open the OpenClaw app expecting it to boot a Gateway. Leave a Gateway running on a machine that is actually on, or accept that the agent is off until that machine is.

The short version, if you are choosing before you leave the house: Grok Bot starts on the phone. A Dot starts on a desktop and can be messaged from the phone afterward. OpenClaw starts on the host that runs the Gateway, and the phone only joins it.