What is ChatGPT Dots? A plain explanation

A dot is an always-on agent inside ChatGPT: an assistant that keeps working after you stop typing. You hand it a job once. It works on that job between your conversations, on its own computer in the cloud, and comes back to you with results and with the decisions that need you.

#What does a dot look like in ChatGPT?

A dot lives in the ChatGPT you already use. Look at the left of this screen: this dot is called Zozo, and it sits right under New chat.

ChatGPT with the sidebar open: New chat, then a dot named Zozo, then Projects and Recents. In the conversation the dot has written: Hey! I'm your dot. Message or call me anytime. I'll keep things moving, even when we're not talking, and check in with updates or questions. — Image: Zinho Automates

#What is a dot made of?

Behind that one sidebar entry, a dot is a model with four things attached: a computer, your apps, its own notes, and a line to you.

Everything inside the dashed box runs in the cloud. That's why closing your laptop changes nothing.

Here's what each part gives your dot.

Part	What it gives your dot
A model	GPT-6 Astra does the reading, planning and writing.
Its own computer	A cloud computer with a browser and files, where your dot researches, builds and keeps going while your devices are off. The dot's cloud computer
Your apps	Email, calendar, files and more, through the plugins you've connected to ChatGPT and within the permissions you've set. Apps and permissions
Its notes	Your dot's own saved notes on your preferences, decisions and ongoing work, on top of what ChatGPT remembers. Dot memory
A line to you	ChatGPT on your computer or phone, a voice call, or Slack. It's the same dot everywhere, with the same memory.

Your own computer isn't on the list. A dot can use it too, but only if you allow it, and that's off until you do. Let your dot use your computer

#What does a dot do while you're away?

Those parts are why a dot can be "always-on": the work doesn't stop when you do. Try it: the same request goes to a normal chat and to a dot, and then you leave for the day.

You close the laptop. Then what?

At 9:00 the same request goes to a normal chat and to a dot. Drag the day forward and watch what each one does while you're in meetings.

Time	You	A normal chat	Your dot
9:00 AM	Send the same request to both: keep the offsite on track.	Answers with a list and draft replies.	Says what it will do and starts reading.
9:20 AM	Close the laptop and go into meetings.	Waits for your next message.	Reads the plan and the venue emails.
10:30 AM	In meetings.	Nothing.	Delivers the list: 3 decisions, 2 deadlines, 1 thing waiting on the venue.
12:15 PM	In meetings. The question reaches you on your phone.	Nothing.	Sees the headcount go to 26, updates the list, drafts a note to the venue and asks before sending it.
2:30 PM	In meetings.	Nothing.	Tells you about a calendar clash you never asked it to look for. It changes nothing.
4:00 PM	In meetings.	Nothing.	Needs the venue's site, which needs your login. It hands that step to you.
5:30 PM	Open the laptop.	Still shows its 9:00 answer, with the old headcount.	Tells you what is done and what is waiting for you.

The email at 12:15 depends on you. Say yes and the note goes to the venue. Say not yet, or say nothing, and the draft waits. Nothing is sent.

you laptop closed

a normal chat waiting for your next message

your dot

9:00 AM

You're at your desk, and you send the same request to both.

A normal chat

Answers within a minute: a list and draft replies.

Your dot

on its ownSays what it will do, then starts reading.

A normal chat

ChatGPT

Help me keep our offsite on track. Read the plan and the venue emails, list what needs deciding this week, and draft replies for me to review. Don't send anything.

Here's what I found in the plan and the venue emails: 3 decisions, 2 deadlines, and 1 thing you're waiting on from the venue. Draft replies are below. Ask me again when something changes.

Ask ChatGPT

Your dot

dotThinking...

Help me keep our offsite on track. Read the plan and the venue emails, list what needs deciding this week, and draft replies for me to review. Don't send anything.

Read 9:00 AM

On it. I'll go through the plan and the venue emails and list what needs a decision this week. Drafts come to you first, and I won't send anything.

Review offsite plan

10:30 AMdid this on its own

Your list is ready: 3 decisions, 2 deadlines, and 1 thing we're waiting on from the venue. Catering is due Thursday, so it's first.

Offsite: what needs you this week

3 decisions · 2 deadlines · 1 waiting on the venue

12:15 PMstops to ask you

Two more people confirmed, so you're at 26. The venue's quote covers 24. I updated the list and drafted a note asking them to price it for 26.

May I send it to the venue?

2:30 PMnoticed without being asked

Something you didn't ask about: tomorrow's budget review moved to 3:00, the same time as your dentist. I haven't changed anything.

4:00 PMhands a step to you

The seating plan is in your account on the venue's site. I can't open it without your login.

5:30 PMyou're back

I'm back. What needs me?

Send a message

When your dot asks its question at 12:15, answer it, or keep dragging and see what happens to the email. Open on its own page

A dot does three kinds of things without a new message from you:

Work you gave it. Your dot decides when to pick the job up again. You don't have to schedule every follow-up.
Work on a schedule. Ask for a check every weekday at 9:00 and your dot saves it and runs it. Give your dot a job
Looking for ways to help. Your dot reads the apps you've connected and tells you what it finds. This reading can't send messages, change anything in your apps, or control a browser.

#When does a dot stop and ask?

A dot working alone still stops at certain points. Before your dot does anything that could affect your accounts or share information, an automatic review checks the action against your instructions and OpenAI's safety rules. Three things can come out of that review:

Outcome	What falls under it
Your dot goes ahead	Reading what you've connected, research, analysis, writing drafts.
Your dot asks you first	Sending a message, sharing a file, buying with a card saved on a shop's site, permanently deleting data, installing software from an unrecognized source.
Your dot hands it to you	Changing a password, moving money, typing a website login.

Asking looks like this: a plain message in the conversation. You answer in your own words.

A dot writes: Gulf Air's site requires a CAPTCHA before I can verify its final fare rules. May I complete this CAPTCHA? You can also give me permission to handle future CAPTCHAs if you'd like. The owner replies: Not today, don't book anything. Just keep the comparison. The dot answers: Understood, no booking. I'll keep this to the three-flight comparison. — Image: Zinho Automates

Approving one message doesn't allow the next. To change what your dot asks about, set a custom rule.

#Do you need a chat, a scheduled task or a dot?

With all that in mind, not every job needs a dot. Ask what happens after the first answer: is the job finished, does it repeat unchanged, or does it keep changing?

Sort it

Sort each job into the least you need for it.

Send me today's meetings every weekday at 8:00The same job at the same time. A saved prompt on a schedule covers it.
Remind me on the first of each month to send invoicesFixed time, fixed job, nothing to decide between runs.
Rewrite this email so it sounds less annoyedOne question, one answer, and you're done.
Keep our offsite on track for the next month and tell me when a deadline is at riskThe job changes from day to day, and something has to notice and follow up.
Keep my sales proposal current as the customer's requirements changeNobody knows when the next change comes. A dot watches and brings you the decisions.
Explain a clause in my leaseNothing to follow up once you've understood it.

A chat is enough

A scheduled task

A dot

Show the answer

A chat is enough: Rewrite this email so it sounds less annoyed (One question, one answer, and you're done.); Explain a clause in my lease (Nothing to follow up once you've understood it.)

A scheduled task: Send me today's meetings every weekday at 8:00 (The same job at the same time. A saved prompt on a schedule covers it.); Remind me on the first of each month to send invoices (Fixed time, fixed job, nothing to decide between runs.)

A dot: Keep our offsite on track for the next month and tell me when a deadline is at risk (The job changes from day to day, and something has to notice and follow up.); Keep my sales proposal current as the customer's requirements change (Nobody knows when the next change comes. A dot watches and brings you the decisions.)

Scheduled tasks come with ChatGPT Go, Plus and Pro, so you don't need a dot for those two. Dots, Codex, Work and scheduled tasks compared

#Next step

You know what a dot is. Next, find out if your account can have one.

Next lessonCan I get Dots?

Check if you can get a dot in three taps: which ChatGPT plans include one, where Pro leaves it out, and what a work account needs before Your dot appears.

Who can get a dot?

Dots come with ChatGPT Pro outside the European Economic Area, Switzerland and the UK, with a Business Premium seat, and in an Enterprise workspace where an admin has turned dots on. You have to be 18 or over. Check yours

Does talking to my dot use up my ChatGPT limits?

No. Conversations with your dot don't count toward your ChatGPT usage limits. Tasks your dot starts in ChatGPT Work or Codex count toward those limits as usual. Dots usage limits

Is a dot the same as ChatGPT Work or Codex?

No. Work and Codex do one job at a time, when you ask. A dot keeps a job going and can hand pieces of it to Work or Codex. Dots, Codex, Work and scheduled tasks compared

Can a dot make mistakes?

Yes. Review anything that matters before you rely on it. Stopping a task doesn't undo what your dot has already done.

What are specialist dots?

A separate preview for organizations. A specialist dot takes on a dedicated job inside a company, with its own identity, credentials and access to the systems it needs. Specialist dots are available only through pilots run with OpenAI's engineering teams.

6 October 2026
Rebuilt around real screens and a simulator of a day with a dot.
5 October 2026
First published.