List

ChatGPT dots: what to set up before you turn one on

The short version

Before you turn a dot on: start with one job you already check often, set the rule to "ask first" for anything that messages people, connect only the apps that job needs, and remember that password changes and money transfers always stay with you. This is OpenAI's description of its own product, read on 11 October 2026, not a test of mine.

By Wlad Kowalski and the Mellow team · Updated October 14, 2026, sources read October 11, 2026

What a dot is

A dot is an always-on agent inside ChatGPT, announced by OpenAI on 29 September 2026 and powered by GPT-6 Astra. It has its own computer and browser in the cloud, keeps working between conversations and while your computer is off, connects to your apps through plugins and writes to you in ChatGPT, Slack or Teams.

Before you turn it on

Start with one job you already check often. Share its sources, review the first result, then widen it.
Add Custom rules in Settings → Personalization. Each action gets one of four: act without asking, act when you say so, ask first, hand off to you. Use "ask first" for anything that messages people.
Connect only the apps that job needs. It reads them on its own and keeps notes. Disconnecting doesn't erase them, and you can't delete one memory, only the whole dot.
Password changes and money transfers always stay with you. Purchases need your approval.
Watch Activity in the desktop app. Pause stops the main task only, and stopping doesn't undo what's done.

Plans and regions

The first dot is included in ChatGPT Pro (from $100 a month) and Business Premium ($125 per user a month, or $100 billed annually; Business needs two or more users).
Free, Go ($8) and Plus ($20) don't include it.
Pro: users over 18 outside the European Economic Area, the United Kingdom and Switzerland. Business Premium: all supported ChatGPT regions. Rolling out gradually, and access may take several days.
Conversations with a dot don't count toward ChatGPT usage limits; tasks it starts in ChatGPT Work or Codex do.

What OpenAI says about mistakes

In OpenAI's own test of a long chain of tasks (ten tasks between the first and the last), 19.7% of samples were flagged for moderate scope violations, meaning actions that exceeded the intended task; with five tasks the rate was 8.6%, and there were no severe breaches. In another test OpenAI sent 50,000 emails, 16,600 of them attacks, with no scored attack successes. OpenAI's Help Center still says the protections reduce the risk but do not eliminate it.

Questions people also ask

Which plans include a dot?

ChatGPT Pro, from $100 a month, and Business Premium. Free, Go and Plus don't include it. Enterprise workspaces can switch on a beta that the workspace admin must enable.

Is a dot available in the EU and the UK?

Not on Pro yet: Pro access is for users outside the European Economic Area, the United Kingdom and Switzerland. Business Premium is available in all supported ChatGPT regions. The list says "yet", so check it on the day you buy.

Can I delete one thing a dot remembers?

No. OpenAI's Help Center says you currently cannot view, delete or directly modify individual dot memories; to delete its context you delete (reset) the dot. Disconnecting an app doesn't delete what the dot already learned from it.

What does "ask first" do?

It's one of four settings for each action in Custom rules. With it the dot asks you before acting, which OpenAI's own example uses for messages to customers.

Sources

Other companies’ features and prices change. Every statement about them rests on these pages, as they read on the date above.