Does an OpenAI Dot keep working after you close ChatGPT?

Cloud work on an OpenAI Dot keeps going after you close ChatGPT. OpenAI’s getting-started page calls a Dot an always-on agent that can keep making progress between conversations, and the computers guide says the Dot lives in the cloud, stays available when your computer is off, and can continue cloud work while your devices are off. The exception is your own computer. If you connected it, that machine has to stay online and the ChatGPT app has to stay open while the Dot uses it. Closing the app makes that computer unavailable. It does not revoke the connection, and it does not stop the Dot’s own cloud computer.

Dots were announced on 29 September 2026. OpenAI says they are powered by GPT-6 Astra, each with its own cloud computer and browser, and that the first one is included for ChatGPT Pro and Business Premium in eligible markets, with an Enterprise beta when an admin turns it on. You create the first Dot in the desktop app or a desktop browser. After that you can message it from mobile. Closing the window you used to set it up is not the same as turning the Dot off.

What keeps running when every screen is dark?

The cloud computer. It has its own files, software, and browser sessions. A website session can remain until you sign out or the site expires it. Those sessions are the Dot’s. Signing in on your personal browser does not sign the Dot in, and the Dot’s cloud browser does not inherit your logins. Cloud work can sit there between periods of use and resume with that state. You do not have to leave a ChatGPT tab open to keep that machine alive.

The Dot can start separate cloud threads. They show up in the desktop, web, and mobile apps so you can review results and add instructions while it keeps coordinating. You can talk to it while several threads run. For coding, it can use a Codex cloud environment you already configured. None of that coordination is a process inside the ChatGPT window on your laptop. The window is a view. Activity, in the desktop app, is where you open the Dot’s profile and look at a task’s progress, files, and results, including work that happened in the background. If a task is waiting on a decision, an app connection, a sign-in, or an approval, you open that request. Closing ChatGPT pauses your ability to answer. It does not, by itself, cancel the cloud thread.

You can message the same Dot from the phone while it works in the cloud. Contact channels are separate from computer access. Adding Slack, or a texting path where that is offered, does not connect your inbox and does not require the desktop app to stay open for cloud work. Desktop is where OpenAI tells you to connect those messaging channels. Once they exist, the phone is a legitimate place to send the next request. I have not found a help page that promises every scheduled task survives an account reset. Reset is a different act: it deletes the Dot, including conversations, saved memories, and scheduled tasks. Closing the app is not Reset.

What stops when you close the app?

Work that was using your computer. Local access starts turned off. You connect one personal computer at a time, from the ChatGPT app on that computer: the Dot’s profile, Computers, Your computer, Allow access, then confirm Allow access. The cloud computer stays connected either way. After you allow it, the Dot can use local files and apps from any channel you use to message that same Dot, including the phone. The connection persists between tasks. While the Dot is actually using the machine, OpenAI says to keep the computer online and the ChatGPT app open.

An Offline status means the computer is unavailable. It does not revoke access. Revoke access is the control that removes it. Sleeping the laptop, quitting ChatGPT, or losing the network produces unavailability. The next cloud task does not automatically become a local task to compensate. If a site blocks the cloud browser, you can ask the Dot to try the connected computer, and that may create a separate local task. It does not move the cloud browser’s session onto your machine, and it does not guarantee a particular local browser. If the app is closed, that fallback has nowhere to run.

Local skills need a connected computer. A task can be managed in the cloud while using files on the machine you connected. Open the task if you need to see where the work is happening. When you close ChatGPT on that machine, you should expect the local half to stall and the cloud half to be the part that can still move. Codex or Work Sync connections are separate. Neither one, by itself, gives the Dot your computer. The Allow access switch is the one that does.

What still waits, even though the cloud computer is up?

Anything that needs you. Before a Dot takes an action that could affect accounts or share information, an automatic review checks it against your instructions, plugin permissions, custom rules, and built-in safety requirements. The review can let it proceed, ask for approval, or hand you a step you must do yourself. Changing a password is the example OpenAI gives for a step you must do. The most sensitive actions, including changing a password or transferring money, require you to take over. Other actions, such as permanently deleting data or installing software, may require approval each time. Custom rules can tell the Dot which supported actions it may take alone, which need you, and which it should not take. They cannot turn off those core safety requirements, and they cannot change the separate safety review. A rule is an instruction the Dot tries to follow. OpenAI says it can make mistakes. A rule also does not grant access to an app or a computer, and it does not remove the confirmation to use a saved login.

Sign-in is the same kind of pause. For supported flows, the Dot sends a private form. You enter credentials there, and any verification code the site wants. The form sends them to the remote browser outside the conversation. The privacy FAQ says that path does not expose the credentials to the model. You can instead take over the Dot’s browser and sign in on the site yourself, then return control. Save to Passwords is optional. Reusing a saved login to sign in again requires your confirmation. An already valid session does not need that confirmation. Closing ChatGPT does not type the password for you, and it does not keep a takeover session open on a machine you shut.

Plugins follow the permissions you already set for ChatGPT. You can allow a read without allowing a send. If a connection expires, the Dot has to be reconnected before it continues. Those permissions apply across Dots, ChatGPT, ChatGPT Work, and Codex. Disconnecting an app does not delete information the Dot already obtained. Deleting that information means deleting the Dot. Closing the window deletes nothing.

How do you tell the two computers apart after you reopen the app?

Open the Dot’s profile and look under Computers. The cloud computer is the one you can inspect, and Open computer on a browser handoff shows it to you. Opening it does not give you the mouse. Take over does, and Return control gives it back. Your computer is the optional row, with Allow access or Revoke access, and an Offline state when the app is not there to serve it. Activity lists tasks, including background ones. A task that names your computer is the one that needed the app open. A cloud thread is the one that could move while the laptop was shut.

Progress between conversations is the phrase on the getting-started page, and it is narrower than “the Dot will invent new projects all night.” It can keep going on work it already has, and the controls page says a Dot can also research using tools it already has. Proactive research has restrictions custom rules cannot lift. I am not going to describe those restrictions as a blank check to wander. When you reopen ChatGPT, Activity is the list of what moved and what is blocked on you. Answer the blocked items. Do not treat a green cloud computer as proof that a local file task finished.

The September announcement also says a plan includes an allowance for deeper work, with extended limits for the first month. I have not verified a dollar figure or a promise that Dot usage never touches your ChatGPT allowance, because the help pages I read talk about an allowance and about extended launch limits rather than a permanent exemption. Closing the app does not pause that allowance. A cloud job that keeps running is still work. If you are near a limit, quitting the window is not how you stop the meter. Pause the task, or stop giving it new ones, from the Dot itself.

If you only remember one test, use this. Give the Dot a cloud-only job, quit ChatGPT, wait, and reopen Activity. The computers page says that work can continue while your devices are off. Then repeat with a job that needs a file on the connected computer, quit the app, and expect Offline rather than progress. The Dot did not stop existing. The half of the work that lived on your machine did.