Can ChatGPT Dots Track Your Restaurant Opening While Your Computer Is Off?

Yes, according to OpenAI's documentation. A ChatGPT dot works on its own cloud computer, so it can keep a tracker built from notes you've given it while your laptop is shut. Anything that lives on your own computer is different: the dot can use it only while that computer is connected, online and has the ChatGPT app open. The part that doesn't change is that a dot can't tell whether a date is real. The owner still confirms every delivery date, makes the soft-opening call and approves anything that leaves the tracker.

This is a documentation-based workflow guide, not a review. We read OpenAI's dots documentation on October 1, 2026. We haven't used a dot on a real opening, and the prompt below hasn't been tested. Juniper Counter, its dates, notes and people are fictional, and the tracker shown is an illustrative example of the structure to expect. It isn't output from a dot.

Who can use a dot right now

OpenAI added dots in its weekly update for September 28 to October 2, 2026, and says they are rolling out gradually. That page gives a release week, not an exact launch day. As of October 1, the access section of the dots documentation lists:

• Pro 100, Pro 200 and Pro 500: users over 18 outside the European Economic Area, the United Kingdom and Switzerland.
• Business Premium: rolling out worldwide.
• Enterprise: rolling out worldwide, off by default until a workspace administrator turns it on.

Other plans aren't on that list. Being on an eligible plan doesn't mean you have it yet; OpenAI says you may not see dots immediately. Open ChatGPT and look before you plan around it.

You create a dot in the ChatGPT desktop app or in ChatGPT in a desktop browser. After setup you can talk to the same dot in the mobile app once the supporting update reaches you. Mobile web isn't supported.

On usage, the documentation says conversations with your dot don't count toward your ChatGPT usage limits, but tasks it starts or manages in ChatGPT Work or Codex count toward those products' usage limits as usual. Your plan also includes an allowance for deeper work, with extended limits for the first month after launch. The pages we read don't give the size of that allowance, so don't treat a dot as unlimited work.

What runs with your computer off, and what doesn't

OpenAI's documentation keeps three kinds of connection apart, and it helps to do the same:

Two more details matter in practice. Connecting Slack to talk to your dot doesn't give it access to your inbox, other apps or your computer. And the dot's cloud browser has its own sign-ins: being logged in to an equipment dealer's portal on your laptop doesn't log the dot in.

The example: three notes from a fictional opening

Juniper Counter is an invented counter-service restaurant opening to the public on Monday, November 9, 2026. That date belongs to the example; it has nothing to do with the product. The owner has three notes:

1. Equipment dealer, Tuesday, October 6: "The combi oven and the two-door reach-in are on the freight schedule for the week of October 26. We'll confirm the exact day once the carrier books it. Install and startup are done by our tech after delivery."
2. General manager, Thursday, October 8: "Line training is four days hands-on plus two mock services, six days in a row. It needs the combi oven installed and working. I can't schedule it until the equipment is in."
3. Owner, Friday, October 9: "Public opening is Monday, November 9. That's set. Friends-and-family soft opening on November 5 or November 6. I'll decide once the equipment date is confirmed."

The trap is in note 1. "The week of October 26" is a window and a promise to confirm later. A tracker that turns it into "Delivery: October 26" has made up a deadline, and the training plan and soft opening would quietly get built on it.

What the tracker should look like

Here is an illustrative target, as of Wednesday, October 14. Every row points to the note it came from, and anything not confirmed in a note stays Unknown.

The Stale flag comes from a rule the owner set for this example: any Unknown row whose newest note is more than seven days old gets flagged. October 6 to October 14 is eight days, so the dealer row is stale. The GM's note is six days old and the owner's is five, so neither is flagged.

Check the dates yourself

The useful thing a tracker can add is arithmetic you can check. Under the GM's plan of six training days in a row, every day including the weekend:

• Soft opening Thursday, Nov 5: training has to run Friday, Oct 30 through Wednesday, Nov 4. The combi oven has to be installed and working by the end of Thursday, Oct 29.
• Soft opening Friday, Nov 6: training runs Saturday, Oct 31 through Thursday, Nov 5. The oven has to be working by the end of Friday, Oct 30.
• The dealer's window: if "the week of October 26" means Monday, Oct 26 through Friday, Oct 30, a Friday delivery can't support a Nov 5 soft opening at all. It supports Nov 6 only if install and startup finish the same day.
• Days to fix problems before Nov 9: three days after a Nov 5 soft opening (Nov 6, 7 and 8), two after Nov 6 (Nov 7 and 8).

Those are calculated dates, not confirmed ones. Each depends on assumptions: that the dealer's week runs Monday to Friday, that training runs every day, and that nothing else blocks the line. A tracker should label them as calculated and show the assumptions, so nobody reads them as commitments.

The prompt

Paste the notes yourself. This version doesn't ask the dot to connect anything, start other tasks, contact anyone or keep checking on its own.

Help me keep a draft opening readiness tracker for my restaurant. Use only the notes below. Don't use other sources, apps or files, and don't connect anything new. Work in this conversation only: don't create Work or Codex tasks, separate threads or scheduled checks.

Build a table with these columns: Task, Supplied evidence and date, Owner, Depends on, Status, Next question.

Rules:
- Status is Confirmed only when a note states the fact or date as settled. Anything described as a window, an estimate, planned, tentative or "to be confirmed" is Unknown.
- Never turn a window or an estimate into a specific date or deadline.
- Mark any Unknown row as Stale when its newest note is more than 7 days old. Today is [date].
- If you calculate a date, label it Calculated and show the arithmetic and the assumptions under the table.
- Keep what the notes say separate from what you infer.

After the table, list the decisions only I can make, and draft the questions I need to ask each person. Label the whole thing DRAFT, NOT A SCHEDULE.

Don't send, post or share anything. Don't book, buy, schedule or change anything, including payroll, staff schedules and our website. Don't contact anyone. Show me the result and wait for my corrections.

NOTES:
[paste your notes, each with who wrote it and the date]

Check the first result against the notes line by line before you rely on it, which is also what OpenAI's getting-started guide suggests. Look hardest at three things: no window became a date, every Confirmed row names a note that actually says it, and the calculated dates match the arithmetic above.

Decisions only the owner makes

• Soft opening date, Nov 5 or Nov 6, and the latest day you'll make that call.
• What happens if the oven is late: move the soft opening, change the training plan with the GM, or move the public opening. The tracker can lay out the dates for each option. It shouldn't pick one.
• Who sees the tracker. A draft with Unknown rows shouldn't go to staff or vendors as if it were a schedule.
• What the dot may do on its own. In this setup, nothing beyond updating the draft and drafting questions for you.

Keep the draft from becoming a commitment

• Training shifts go on the schedule only after the GM confirms them and the equipment date is confirmed by the dealer, not by the tracker.
• Nobody gets a date from the tracker until the person responsible for it has confirmed it in writing.
• Drafting a message isn't permission to send it. OpenAI's controls page says the same thing. Keep "don't send" in your instructions anyway.
• A row about a permit or an inspection records what you have in hand. The tracker isn't a compliance check; the permit or inspection report is the evidence.
• Keep employee pay, personal details and guest information out of it. The example doesn't need any of that.

If you'd like a stronger guardrail, OpenAI documents optional custom rules under Settings > Personalization > Custom rules, with choices such as Ask before taking action or Hand off to you. OpenAI describes them as instructions the dot tries to follow and says it can make mistakes, so they don't replace reading the result.

Updating it later

When new information comes in, go back to the same conversation. The documentation says you can return with new information or priorities without restarting the work. Say the dealer writes on Thursday, October 15: "Delivery and install Thursday, October 29, tech on site 8 a.m. to 2 p.m., startup check the same day." Paste it in and ask:

New note from the equipment dealer, October 15: [paste it]. Update only the rows this note affects. Keep the earlier evidence in the row and add the new one. Tell me which statuses changed and why, recalculate any Calculated dates and list the questions that are still open. Don't change anything else and don't contact anyone.

A good update moves delivery to Confirmed for October 29 and leaves "installed and working" Unknown until the startup check is done. It should also point out that a Nov 5 soft opening now works only if startup passes that day, with no slack. That's a decision for the owner, not a status change.

If you want updates at a fixed time, OpenAI's documentation says to give the time, time zone and an end date, ask the dot to confirm the schedule, and review it under Scheduled. We'd start with updates you ask for and add a schedule only once the first few results check out. A schedule isn't a guarantee that anything gets checked; read what it reports.

How to stop it

Stopping takes more than one step. According to OpenAI's controls page:

• Pause stops the dot's current main task. It doesn't stop delegated tasks or cancel scheduled runs.
• Open a delegated task in Activity to stop that task.
• Open Scheduled to disable or delete a recurring task.

Ending a conversation or a voice call doesn't necessarily stop assigned work. Stopping doesn't undo anything already done, and deleting a dot doesn't undo changes in connected apps or recall messages already delivered.

Limits

This guide is based on OpenAI's documentation as read on October 1, 2026, for a feature that is still rolling out. Availability, limits and controls can change, so check the current pages before you set anything up. We haven't tested dots or this prompt, and we don't know how well a dot follows these instructions in practice. OpenAI says your dot can make mistakes and that a completed run doesn't by itself confirm the result. Juniper Counter and every date, note and person in it are fictional. A tracker, by AI or by hand, doesn't confirm a delivery, certify a permit or inspection, or make a decision for you.

For the opening plan itself, see the free restaurant opening planner. For training against observable standards, see hiring and training your opening team, and for what the soft opening should test, how to run a soft opening that finds problems. If you already run ChatGPT scheduled tasks, the register in our GPT-5.5 automation checklist is a good place to record a dot's schedules too.

Sources

• OpenAI, ChatGPT documentation, "Meet dots," including the Access section, read October 1, 2026: Meet dots
• OpenAI, ChatGPT documentation, "September 28–October 2, 2026," Delegate ongoing work to your dot, read October 1, 2026: What's new, September 28–October 2
• OpenAI, ChatGPT documentation, "Connect computers and apps to your dot," read October 1, 2026: Computers and apps
• OpenAI, ChatGPT documentation, "Tasks and memory," read October 1, 2026: Tasks and memory
• OpenAI, ChatGPT documentation, "Control your dot," read October 1, 2026: Controls
• OpenAI, ChatGPT documentation, "Get started with your dot," read October 1, 2026: Getting started