OpenAI began rolling out Dots on September 29, 2026: agents that can continue tasks between conversations, use connected apps and work on a separate cloud computer. The practical decision for a new user is what the Dot may read, what it may do without another request and how to inspect its work. This guide follows OpenAI’s current setup instructions and privacy and safety FAQ, checked September 29. BIG CHANGE has not created or tested a Dot; the steps and controls below are documented by OpenAI.

The big change

  • What changed: A ChatGPT user can assign an ongoing goal to a Dot and let it work through connected apps between conversations. The user chooses those connections and can set rules for actions.
  • Why it matters: An app connection gives the Dot a continuing route to information. OpenAI says disconnecting it ends new access but does not remove information the Dot already absorbed. The connection deserves a decision before setup, not only when work is finished.
  • What to watch: The visible record is the Dot’s Activity View, where a user can inspect steps, correct direction or ask it to stop. Whether a completed action can be undone depends on the action and app, according to OpenAI.

Check access before starting

According to OpenAI’s launch post and setup article, the first Dot is included with ChatGPT Pro or Business Premium at no extra cost. Pro rollout excludes the European Economic Area, Switzerland and the UK; Business Premium covers supported ChatGPT regions. Enterprise access is a beta that a workspace administrator must enable and is off by default. The rollout is gradual, so eligible accounts may wait several days. Dots are unavailable to people under 18.

Create a Dot in the ChatGPT desktop app or on desktop web. OpenAI says creation is unavailable in the mobile app or mobile web; after setup, you can talk to your Dot in the mobile app where access is available. You need an eligible account, a desktop session and a task you can describe. If an Enterprise workspace does not show Dots, the administrator’s enablement is the first thing to check. This article does not establish access for any particular account.

Create the Dot and choose one app

Open the Dot setup in ChatGPT on desktop and follow the onboarding prompts. You can give it a name. OpenAI’s setup article says you can connect Slack and a personal email account, while messaging channels are set up from desktop. A personal email connection does not give the Dot its own standalone email address.

Before connecting anything, inspect permissions in ChatGPT’s Plugins tab. OpenAI says they are shared across Dots, ChatGPT, ChatGPT Work and Codex, so an existing connection may already be available to the Dot. Connect only an app needed for the first task, and review what access that app requests. This is our setup recommendation.

The Dot has its own cloud computer. Access to your local computer starts off. Connecting one through the desktop app and confirming Allow access can let the Dot use local files and work on that computer from any of its messaging channels. OpenAI says Revoke access stops that access. Leave the local connection off unless the task requires it.

Give it a bounded task and action rule

Tell the Dot the goal, which app to use, the result you want and which actions need review. A bounded instruction could be:

Text
Use the connected calendar to list tomorrow’s meetings and return a summary. Do not send invitations or change events.

This illustrates an instruction; BIG CHANGE has not run it. OpenAI says you can revise a request in the conversation.

Open Custom Rules and describe the action the rule covers. OpenAI gives the mobile path as the Dot’s profile → Customize → Custom rules. Choose Take action without asking, Take action if pre-approved, Ask before taking action, or Hand off to you. “Pre-approved” means explicitly requested in your prompt. Save with the checkmark and review the rule. For email or calendar changes, Ask before taking action configures the Dot to request approval; OpenAI warns it can make mistakes following rules. Some sensitive actions require you to take over regardless of your rule.

OpenAI describes separate proactive research that can read permitted connected information in the background and make private notes. Its research tools cannot directly send messages, change plugin content or control a browser or computer. That limit does not make every Dot task read-only: previously authorized tasks can continue under their action rules and safety checks. OpenAI describes its safeguards; BIG CHANGE has not independently measured their effectiveness.

Inspect work, pause and understand deletion

In the desktop app, open the Dot’s profile and check In progress, Scheduled and Completed. OpenAI says Activity View shows task descriptions, status and steps; you can add context, correct a misunderstanding, change direction or ask the Dot to stop. Review the actual result before relying on consequential work. OpenAI warns that some completed actions cannot be undone.

To stop ongoing work until you resume it, use the profile’s ••• menu and select Pause. The profile then offers Paused • Tap to resume. For a local computer connection, Revoke access is the separate control. To stop new access to a connected app, disconnect it in Plugins. None of these should be confused with erasing the Dot’s existing context.

OpenAI says you cannot inspect, edit or delete individual Dot memories. Reset in the profile menu deletes the Dot, its conversations, saved memories and scheduled tasks. It does not delete files, Codex threads or ChatGPT conversations the Dot created elsewhere; those follow their own storage and retention controls. Information shared into regular ChatGPT memory also has separate controls. Review the deletion notice before confirming Reset.

OpenAI says Dot usage does not count toward eligible Pro, Business and Enterprise plan allowances for the next month. It has not yet announced the terms after that period. OpenAI also says tasks a Dot starts or manages in Codex or ChatGPT Work count toward your usage limits as usual. Texting is a limited US Pro beta with possible carrier message and data charges; it is not required for this setup.

Sources & further reading