OpenAI Dots Explained: Plans, Availability in Europe and a Setup Guide for Business

On 29 September 2026 OpenAI showed Dots on stage in San Francisco: AI agents inside ChatGPT, each with its own cloud computer and browser, that keep working on your tasks while you are away. A Dutch business owner can get one through Business Premium. ChatGPT Pro, which also includes a dot, is not offered in the EEA (stand 15 October 2026).

I went through what OpenAI and the press have published since that day, and this article is the result. It was last updated on 15 October 2026. Prices and availability change fast, so I put a date next to every fact.

One thing up front. I have not used Dots myself. It is not available yet for a personal licence in the Netherlands, so everything about Dots here comes from OpenAI and from other people's tests. I do use Grok Bot, a similar tool, and I will get to that.

Dots at a glance

What is a dot, really?

Picture the keynote. OpenAI's yearly developer event, DevDay, brought more than 20 announcements, according to the DevDay recap, and ChatGPT has 1.2 billion weekly users. TechCrunch and NBC News wrote about it the same day. NBC adds that a live demo failed on stage.

A dot is an AI agent. That is software which gets a goal, works out the steps and carries them out in other programs. Until now ChatGPT answered questions. A dot logs into your apps, opens websites in its own browser and finishes tasks on a schedule. Close the chat and it keeps going.

You start with one primary dot. You give it a name, which becomes its handle (say @shop-dot), and you pick a character or a pet that shows up on screen. OpenAI says teams of several dots are a plan for later. Today you get one.

About the model: OpenAI says Dots runs on GPT-6 Astra. NBC and CBS use other names in their coverage, and one of them mentions a version called GPT-6.1 Astra that OpenAI has held back.

What does it do all day?

Every dot sits on a cloud computer with a browser. It reaches your tools through plugins, small connectors between the dot and an app. OpenAI and NBC speak of more than 4,000 apps. Named examples are Slack, Teams, Google Drive, Gmail, Google Calendar, Docs and Sheets, GitHub, Figma, Adobe Creative Cloud, Outlook and Codex. I found no confirmation for Dutch accounting packages such as Moneybird or Exact. If your bookkeeping lives there, check before you count on it.

Your own computer stays out of it unless you connect it, and that is off by default. A dot has no email address of its own at launch, but you can connect your personal mail. Stored passwords are used without the model seeing them. Changing a password stays with you, always.

A dot also remembers. It learns from your feedback and writes its own notes. Disconnecting an app does not erase what the dot already pulled in. Only a reset of the dot deletes conversations, memories and tasks, and you cannot undo a reset.

Recurring tasks live under Scheduled. The dot can also do proactive research, looking things up without being asked, but only inside connected apps and only reading. Conversations with a dot do not count toward your ChatGPT limits. Codex and Work tasks do, and a dot can start Codex tasks, according to an OpenAI community update of 9 and 10 October.

You reach it through ChatGPT on desktop, web and mobile, and through Slack and Teams. You can call your dot by phone. It cannot call you at launch. For text messages the sources disagree: the OpenAI blog says "soon", the help article describes a limited beta for Pro users in the US.

Who holds the leash?

This is the part I would read twice if I ran a company. OpenAI built in rules, and you add your own, called Custom Rules. Each rule picks one of four behaviours: act without asking, act if approved in advance, ask first, or hand the task over to you.

On top of that sits an Auto-review step that checks actions that touch accounts or share information. A draft is no permission to send. Activity View lists what the dot did, and you can pause it at any time.

Which plan do you need?

OpenAI charges no separate fee for the first dot. According to the ChatGPT pricing page it comes with two plans: ChatGPT Pro and ChatGPT Business Premium. Free, Go and Plus do not include a dot. OpenAI places Dots in its higher plans.

For the current prices, including those of Business Premium, go to that pricing page. That is where OpenAI shows its up-to-date terms. I name no amounts here on purpose. Sources contradict each other and the numbers keep moving (stand 15 October 2026). Plans and limits can change too. OpenAI describes teams of several dots as a plan for later, so extra dots and extra capacity may come with their own terms.

The first month after launch has higher usage limits. After that, limits depend on the plan, and OpenAI has published no numbers. Nobody has published what a single task costs either.

And can I use it in the Netherlands?

The answer depends on the plan. Through a business plan: yes. Through a personal Pro licence: no.

The OpenAI help article says Pro is available to users aged 18 and over everywhere except the European Economic Area and the United Kingdom. Switzerland is excluded as well. Business Premium is available in all supported ChatGPT regions. Enterprise, Edu and Healthcare customers get a beta that an administrator has to switch on. It is off by default.

OpenAI gives no reason for the Pro exclusion and no date for a change. It says more users will follow soon. According to NBC, Sam Altman pointed to strict regulation and said he wants to serve Europe.

A Dutch freelancer who wants to start alone cannot buy Pro today. A company can sign up for Business Premium. According to eesel the plan has a minimum of two seats, so check at signup what that means for a one-person company.

What about safety and the GDPR?

OpenAI states that each dot runs on its own cloud computer, has protection against malicious instructions and is monitored, so a dot can be paused or stopped. The company also says dots make mistakes. Data from Business, Enterprise and Edu plans is not used for model training by default.

The GDPR (in Dutch: AVG) question is open. A report on the Enterprise beta by Beri says the beta offers no data or inference residency. That means no automatic guarantee that your data stays in the EU. I found no statement from OpenAI on GDPR compliance. Have a lawyer or a data protection officer look at it before you connect customer data, and get a data processing agreement in place.

Then there is prompt injection. A web page, an email or a document can contain hidden instructions, and an agent that reads it may follow them. OpenAI acknowledges the risk, as NeuralTrust reports. A dot that reads your incoming mail will meet text like that sooner or later.

What the first testers ran into

Every review I found is one or two days old and based on a handful of tasks. Read them as first impressions. The Pro test by AI Agents Library averaged 8.8 out of 10 over ten tasks, with an overall score of 9.2.

• A content plan scored 9.8.
• A branded website built with a custom Skill scored 9.6.
• A weekly report as a PDF scored 8.6, after the tester asked for a PDF explicitly.

The dips are more useful for your planning. Spreadsheet formulas broke in Google Sheets (7.5). Inbox triage scored 7.5 because the dot treated a resolved issue as open. Meeting notes scored 7, because the dot often returned plain text where the tester wanted a draft ready to send.

Google Calendar worked well: the dot could add, move and remove events. Events created through Gmail produced a "Bad Request" error in the OpenAI community forum. Websites, subscriptions and ordering are weak spots, because CAPTCHAs, two-factor authentication and payments block the dot.

The Neuron reports a dot sending an email without the tester noticing, unrequested proactive actions, slow work and a failed phone call. Setup took about five minutes. For invoices, OpenAI itself only claims that a dot can prepare a missed one and send it after your approval.

Where I stand: Grok Bot

Dots is not the only agent that stays on. Our article on Grok Bot explained for business covers the SpaceXAI version, and the head-to-head is in OpenAI Dots vs Grok Bot compared.

Two differences show up in the facts. Grok Bot runs one shared computer per user, hosted in the United States, and has no official region list. Dots is a ChatGPT feature with a clear regional split: Business Premium works in the Netherlands, Pro does not. CBS analyst Crisafulli calls Dots the more business-oriented product of the two.

I use Grok Bot myself from the Netherlands, on a paid Cursor Pro+ plan. I run a team of four little bots: a scout that looks for leads, a stalker that digs up background on them, an analyst that filters and a head of desk that puts the final list together. I let them search for specific leads, with very specific requirements, and when there is a match the stalker picks it up. The end result is a list worth reading. I have no measured results, counts or time savings to share.

If you are weighing Dots, take the method. Give each role one narrow job and let a last step decide what reaches you. A dot is a single agent today, with teams of dots planned for later, so you would split such roles into separate tasks. I have not tried that with Dots, and my Grok Bot setup does not prove it works there.

Sanne and her inbox: a setup you can copy

To make this concrete, take Sanne. She is made up, an example only: she runs an installation company with twelve people. She spends an hour a day sorting email and chasing unpaid invoices and unanswered quotes.

The dot sorts the inbox, writes draft replies, tracks open invoices and quotes, and keeps an eye on the calendar. It sends nothing without Sanne's approval.

What you need:

• A plan that includes a dot: Business Premium in the Netherlands (Pro is not offered in the EEA). Check the current terms on the pricing page.
• A Gmail or Outlook account and your calendar, connected through plugins.
• The desktop app or the desktop web version of ChatGPT for the setup.

The steps follow the OpenAI guide. My comments are mine.

1. Create the dot. Open ChatGPT on desktop, create your dot, pick a name and a character. The Neuron measured about five minutes.
2. Connect your apps. Go to Customize, then Plugins, and connect Gmail or Outlook and your calendar. Leave the connection to your own computer switched off.
3. Tell it about your work. The dot introduces itself and asks what you do. Answer concretely: which clients are active, how fast you reply, which senders always get priority.
4. Set Custom Rules. Use the four behaviours on purpose. Without asking: apply labels, archive newsletters, read your calendar. If approved in advance: move calendar events you pre-approved. Ask first: send any email, accept invitations, create events. Hand over to me: payments, passwords, anything with a CAPTCHA or two-factor code.
5. Give it the first task. Start with one read-only task for a week, then widen it. The prompt below is a starting point.
6. Schedule it and look over its shoulder. Set the repeating task under Scheduled. After the first runs, open Activity View and compare what the dot did with what you expected. Pause it if something looks off.

Every weekday at 8:00, go through my inbox of the last 24 hours. Sort messages into: needs a reply today, waiting for someone else, can be archived. Write a draft reply for every message in the first group in my usual tone: short, direct, no exclamation marks. Save the drafts as drafts, never send them. Then list all invoices and quotes I sent that are more than 14 days old without a reply or payment, with client name, amount and date. Check my calendar for conflicts this week. Send me one summary in this chat, with the drafts listed by subject.

Where it can go wrong:

• Testers saw a dot return text where they wanted a draft that is ready to send. Say in the prompt that the output must be a saved draft.
• One inbox test counted a resolved issue as open. Check the "waiting" list yourself in week one.
• Calendar events created via Gmail returned a "Bad Request" error in the OpenAI community. Create events directly in the calendar plugin.
• CAPTCHA, two-factor codes and payments always stay with you.
• One tester found an email sent without noticing it. Keep sending on "ask first".

I based this setup on OpenAI's guide and on tests by others. I could not try Dots myself: a personal Pro licence is not offered in the EEA, so I have no results of my own to report.

Monday morning, report already there

Same Sanne, second wish. She wants revenue and open invoices in one PDF every Monday, without building it by hand. The dot reads Stripe, a Google Sheet and a Drive folder, writes the report with the figures, the deviations from last week and three actions, and puts the file in Drive.

What you need:

• A plan with a dot (Business Premium in the Netherlands).
• Plugins for Stripe tools, Google Sheets and Google Drive. OpenAI names Stripe tools and Google Sheets among its apps.
• A fixed layout: one sheet with your weekly figures and a Drive folder for reports.

1. Prepare the data. Put your key figures in one Google Sheet with a row per week: revenue, new customers, open invoices, any other number you track. A dot copes better with a plain table than with a sheet full of nested formulas.
2. Connect the apps. Customize, then Plugins: connect Stripe, Google Sheets and Google Drive. Give the dot access to the one folder and the one sheet it needs, nothing wider.
3. Set Custom Rules. Without asking: read Stripe and the sheet, create files in the reports folder. Ask first: edit anything in the sheet, share a file with other people. Hand over to me: refunds and payouts, plus any other payment action.
4. Run the first report by hand. Test the prompt once in the chat before you schedule it.
5. Schedule it. Under Scheduled, set the task to run every Monday at 7:00.
6. Check the first three Mondays. Put the PDF next to your own numbers. Read Activity View to see which sources the dot used.

Create a weekly report for last week (Monday to Sunday). Use the Stripe data and my sheet "Weekly figures". Show revenue, number of payments, new customers and open invoices, each compared with the previous week. Point out every deviation larger than 15 percent and say what probably caused it, based only on the data you can see. End with three concrete actions for this week, ranked by expected effect. Do not change anything in the sheet. Export the report as a PDF of at most two pages and save it in the Drive folder "Reports" under the name "Weekly report YYYY-MM-DD". Tell me in this chat when the file is ready.

Where it can go wrong:

• In the AI Agents Library test, the PDF came after an explicit request. Name the file format in the prompt.
• Spreadsheet formulas written by the dot broke in Google Sheets in that test. Let the dot read the sheet and keep it away from formulas.
• The dot explains deviations from the data it can see. Treat its explanations as guesses.
• Payments and anything behind two-factor stay with you.

The same caveat applies: the steps follow the OpenAI guide and other people's tests. I have not used Dots myself, for the availability reason above, so I have not tested these steps.

What I would do on Monday

Four things matter for a Dutch business owner right now.

Availability. Pro is not offered in the EEA or the UK, and Switzerland is excluded as well. Business Premium works in all supported ChatGPT regions. For a Dutch company, Business Premium or the Enterprise beta are the routes today.

Look up the current prices. OpenAI shows its up-to-date terms for Pro and Business Premium on the pricing page. Check them there before you budget, because prices and limits can change.

GDPR. The Enterprise beta has no EU data residency, according to Beri. So:

• Sign a data processing agreement before you connect customer data.
• Start without customer data: your own calendar, your own reports, public information.
• Ask a lawyer or data protection officer whether you need a data protection impact assessment.

Prompt injection. A dot that reads email or web pages can meet hidden instructions. Keep sending, payments and sharing on "ask first" or "hand over to me", and read Activity View in the first weeks.

Frequently asked questions

Sources

• OpenAI: Introducing Dots (29 September 2026)
• OpenAI Help Center: Getting started with your dot (retrieved 15 October 2026)
• OpenAI: DevDay 2026 recap (29 September 2026)
• ChatGPT pricing (retrieved 15 October 2026)
• TechCrunch: OpenAI launches Dots (29 September 2026)
• NBC News: OpenAI launches Dots, AI agents and safety questions (retrieved 15 October 2026)
• The Neuron: OpenAI Dots hands-on review (retrieved 15 October 2026)
• AI Agents Library: OpenAI Dots review (retrieved 15 October 2026)
• Beri: Dots Enterprise beta, admin controls, data residency and audit gaps (retrieved 15 October 2026)