What Is ChatGPT dot? Features, How to Use It, and Availability

ChatGPT dot is a personal AI agent that continues managing assigned work between conversations. It can research information, prepare documents, track project progress, and return with results or questions that need your judgment. It is useful for responsibilities that require ongoing checks and revisions.

What is ChatGPT dot?

An AI agent organizes work around a goal and uses permitted tools to produce a result. OpenAI calls this feature dots: you create your own dot and give it ongoing responsibilities.

Your dot has a cloud computer and browser. It uses relevant context and your instructions, allowing you to add information or change priorities as work progresses. OpenAI’s introduction to dots identifies GPT-6 Astra as its underlying model. For background on the models themselves, see our GPT-6 vs. GPT-5.6 model comparison.

For example, you could ask it to organize decisions from meeting materials, then check whether new information affects the plan. Define which sources it should use, what it should deliver, and which decisions require your input.

Who can use dots?

As of October 5, 2026, official documentation lists the following eligible plans. Gradual rollout means eligibility does not guarantee immediate access.

Conversations with your dot do not count toward ChatGPT usage limits. Work or Codex tasks it starts or manages use those products’ limits. Plans include an allowance for deeper work; check the limits that apply to your account. See official dot access requirements.

If the menu is missing, check your plan and region. In an organization, also check whether an administrator has enabled access. Eligible accounts may still be waiting for rollout.

What can you do with your dot?

What your dot can accomplish depends on the sources and tools available to it. These scenarios illustrate possible uses, rather than guaranteed outcomes.

Your dot can divide work among background agents and follow up on their results. Each delegated task has its own conversation and receives context for that task; it does not automatically receive every past conversation. See how dot tasks and memory work.

How does dot relate to Chat, scheduled tasks, Work, and Codex?

Choose based on the outcome you need and how the work should continue. A dot manages an ongoing responsibility and can use Work or Codex to carry out parts of it.

Work can handle recurring tasks, and scheduled tasks can return to a conversation with existing context. Persistence alone therefore does not distinguish the features. You can give your dot a continuing role and schedule particular checks within it. The comparison draws on the ChatGPT Work guide and scheduled tasks documentation.

How to create your dot and assign its first task

Start with one small task whose inputs and expected result are clear. Review the output and refine your instructions before expanding its responsibilities.

1. Create your dot on desktop. Open dots in the ChatGPT desktop app or a desktop browser and follow the introduction. You can change its name and appearance later.
2. Connect the apps the task needs. Choose connections during setup or skip them and add them later. A task based on supplied files may not require additional services.
3. Specify the goal and sources. Include the desired result, exclusions, and actions that need your review.
4. Review the result and evidence. Look for missing conditions, misunderstood scope, and figures that differ from the source. Correct those issues before broadening automation.
5. Set follow-up and notification conditions. Specify important changes or a fixed time for results. If you request recurring work, verify the saved schedule.

Initial creation requires desktop. Afterward, the same dot is available in the mobile app when the supporting update is provided. Mobile web is not supported. See the official getting-started guide.

Apps, messaging channels, and your computer: what is the difference?

These connections serve different purposes. Adding a contact method does not grant access to every resource in that service or to your computer.

To connect a computer, open your dot’s profile in the ChatGPT app on that computer, choose Computers → Your computer → Allow access, and confirm. Only one personal computer can be connected at a time. Connecting it to Codex or enabling Work Sync does not automatically grant your dot access.

Cloud work can continue while your computer is off. Steps that need local resources require the computer to be available. Offline does not revoke access; use Revoke access to remove the connection. See computer and app connection requirements.

Slack and Teams reach the same dot, but messages remain in their respective channels. Adding it to a channel does not start monitoring. Specify what to check, when to check it, and when to notify you. See messaging your dot.

How to give clear instructions: practical prompt examples

Include these five elements to make the result easier to assess. The examples below are sample instructions, not reports of completed tests.

• Goal: The problem to solve and the result to produce.
• Sources: The files, pages, or apps to use, with any source restrictions.
• Permitted actions: Whether the work stops at research and drafting or includes actual changes.
• Notifications: When, where, and which changes should trigger an update.
• Output: A reviewable format, such as a comparison table, revisions, or source list.

A one-time article review

Compare the attached blog draft with the official documentation links. Identify dates, product specifications, and access requirements that need verification. Create a table with the draft sentence, official evidence, and suggested revision. Mark anything you cannot verify as unconfirmed. Show me the review only; do not edit the draft file or WordPress post.

This keeps the input and scope manageable. Check that each source actually supports the sentence and add any missing conditions to your next instruction.

A recurring check for official updates

For the next four weeks, compare the official product update pages I specify with my existing article every Monday at 9 AM Korea Standard Time. Notify me in ChatGPT only when a change to features, access requirements, or setup instructions requires an article update. Include the change, source, affected sentence, and proposed revision. Do not publish or edit existing posts. Confirm the saved schedule, execution time, and end date.

Include a time zone and end date, then verify the saved schedule. A scheduled check differs from monitoring an event supported by a connected service. Confirm support before relying on a particular event trigger. See recurring work and event monitoring.

Reviewing work, managing permissions, and stopping tasks

In the desktop app, open Activity in your dot’s profile to inspect delegated work, files, results, and requests for input. A completion indicator does not establish accuracy or delivery; inspect the actual result.

After setup, optional rules are available at Settings → Personalization → Permissions → Custom rules. You can specify actions that may proceed directly, require an explicit request, require approval each time, or must be handed to you. Administrators can disable custom rules.

Rules do not grant separate app access or override built-in safety requirements. Distinguish drafting a reply from sending it, for example. See dot permissions and controls.

Pause does not cancel all delegated work or schedules. Ending a voice call may leave assigned work running. Stopping work does not reverse completed changes.

Frequently asked questions

Does my dot remember every past conversation?

No. It uses relevant ChatGPT memory and its own notes, but those notes are not a complete conversation archive. Changing ChatGPT saved-memory settings does not necessarily change notes your dot has already made. See the official memory explanation.

Does the cloud browser use my personal browser’s signed-in sessions?

No. Its sessions are separate. You can sign in through a private sign-in request or use Take over. Continuing a valid session differs from signing in again with saved credentials; reusing a saved login requires confirmation. See the website sign-in guidance.

Will my dot automatically publish posts if I ask it to help manage my blog?

A request for research or a draft does not authorize publication. Publishing requires an environment with WordPress access, the necessary permissions, and instructions covering that action. Start with source organization and draft preparation, review the results, and then define the scope of further work.