OpenAI Dots: What They Are and Where They Work

👁 134 reads

OpenAI Dots are always-on AI agents in ChatGPT, unveiled at DevDay on September 29, 2026. Each Dot has its own cloud computer, connects to over 4000 applications, and works even when you’re not chatting with it. I checked OpenAI’s help documentation to see who has access: Pro plan users in the European Economic Area won’t get them at launch, as the initial rollout excludes the entire EEA. I haven’t tested Dots myself, so the description below is based on OpenAI documentation and reports from people who had early access.

I cover the full conference in my DevDay 2026 recap. Here’s everything about Dots.

What is a Dot

A normal ChatGPT conversation ends when you close the window. A Dot keeps working. OpenAI describes it as an agent that “takes on ongoing responsibility” and makes progress between conversations. You give it a goal, set what it can do autonomously, and it returns with results and asks about decisions that require your judgment.

Examples from OpenAI’s announcement: for a developer, Dot reviews customer feedback, prepares minor fixes with tests on its own, and delivers completed changes for review. For a content creator, it selects clips from a new interview and writes draft posts. One tester received a reminder from their Dot about an unsent invoice, and after approval, the Dot prepared and sent the invoice.

Each Dot runs on GPT-6 Astra, OpenAI’s most powerful model. Not on the new GPT-6.1 Sol and not on the suspended GPT-6.1 Astra.

How it works: cloud computer and 4000 apps

• Own computer. Each Dot has a cloud machine with Linux and Chrome browser. You can open its screen at any time and see what it’s doing. Your computer is separate from it, unless you connect it yourself in the ChatGPT desktop app (also on Windows).
• Applications. Through ChatGPT plugins, Dot connects to over 4000 applications. Permissions are shared across Dot, ChatGPT, ChatGPT Work, and Codex.
• Where you talk to it. In ChatGPT on desktop, in browser, and on phone, and according to OpenAI’s announcement also in Slack and Microsoft Teams. You can also call it with voice. SMS is in limited beta only for Pro plan in the US.
• Background work. When you’re not talking, Dot conducts “proactive research”: it reads connected sources and takes notes. These tools only work in read-only mode, so they can’t send messages, change content in apps, or control the browser.
• Memory. Dot receives memories from ChatGPT and creates its own, including from connected apps. You can’t review or delete individual memories. The only way is to delete the entire Dot.

You create your first Dot in the desktop app or in a browser on desktop. On phone, you can only talk to it after configuration. The default handle is @yourname-dot, and you can change the name and avatar.

Who has access and regional availability

OpenAI’s announcement only mentions “qualifying markets.” I found the specific list in the help article “Getting started with your dot,” accessed on September 29, 2026.

Additionally, Dots aren’t available to people under 18, and rollout takes “a few days,” so even in an eligible plan, your Dot may appear with a delay. OpenAI hasn’t said when Pro will include the EEA. In practice, the simplest path to Dots in the EEA today is a Business Premium seat in company ChatGPT Business.

Dot pricing and additional Dots

The first Dot is included in the plan price. For the first month, Dot’s work doesn’t consume limits in Pro, Business, and Enterprise plans; after that, OpenAI will announce rules for each plan. Conversations with Dot don’t count toward ChatGPT limits, but tasks that Dot runs in Codex or ChatGPT Work count normally.

About additional Dots, OpenAI only says that “in the future” you’ll be able to purchase additional Dots and increase the speed or monthly scope of work for each one. There’s no price in the announcement, help docs, or ChatGPT pricing page (as of September 29, 2026). Business requires at least two seats, but one can be standard (25 USD). A comparison of all plans is in the article about ChatGPT Pro 500.

Security: Dots and suspended GPT-6.1 Astra

The day before DevDay, OpenAI announced that GPT-6.1 Astra won’t be released because in internal testing it didn’t stick to the scope of tasks and permissions given by the user. This is exactly the capability that agent security with access to email and accounts depends on. Dots run on the earlier GPT-6 Astra, and OpenAI adds several layers of control around the model:

• Auto-review. A separate system checks planned actions, such as sending an email, against the recipient, content, your instructions, and rules. The mechanisms of this control operate outside the environment that Dot can modify, so the agent can’t disable them.
• Custom Rules. For selected actions, you set one of four behaviors: act without asking, act after prior consent, ask before acting, or hand this task to me.
• Hard boundaries. You always change passwords and transfer money between accounts yourself. Permanent data deletion and installing software from unknown sources require consent every time. Purchases with a card saved in a store require approval.
• Passwords. For supported logins, you enter the password in a separate form, and the model doesn’t see it. This doesn’t apply to passwords pasted into chat or a document.
• Monitoring. When the system detects dangerous behavior, it can pause Dot’s work and show a warning.

OpenAI itself admits in the help docs that safeguards against prompt injection from emails and websites “reduce risk but don’t eliminate it.” In my opinion, that’s an honest description. An agent with inbox access reads content written by strangers, and the suspension of GPT-6.1 Astra shows that even the producer can’t always predict when the model will go beyond its assignment. At launch, I’d give Dot read-only access and “ask before acting” rules for everything that goes external. That’s my opinion.

Data: in Business, Enterprise, and Edu plans, OpenAI doesn’t train models on content by default. In personal plans, the “Improve the model for everyone” setting decides, which also covers Dot’s actions and data from connected apps. Even with training disabled, OpenAI employees may in limited cases review Dot activity, such as for security reports.

Early reviews

Dan Shipper from Every tested Dot for several days before launch. He writes that Dot became his main way of using ChatGPT because it filters emails and Slack messages. However, he found the test version too unreliable to recommend: it lost messages, had trouble with permissions and the browser. He advises waiting a week or two. I don’t have my own observations because I haven’t had access to Dots yet.

01OpenAI DevDay 2026: Dots, GPT-6.1 Sol, Pro 500 and MoreOver 20 announcements at OpenAI DevDay 2026: always-on Dots agents, GPT-6.1 Sol at one-fifth the cost of Astra, Ultrafast mode, ChatGPT Space and Pro 500.Read →02GPT-6.1 Sol: Pricing, Availability & Non-English Writing TestGPT-6.1 Sol from DevDay 2026: API pricing in USD, differences from GPT-6 Sol and Astra, availability in ChatGPT plans, and our non-English writing test results.Read →03ChatGPT Space and Pages: What They Are and How to Get StartedUse a touring theatre programme to organise ChatGPT Space, edit a Page, check a corrected date and control who can see related documents.Read →04OpenAI Ultrafast: 300 tokens/s, pricing and supported modelsOpenAI Ultrafast generates up to 300 tokens per second in Codex, 8x faster than standard mode. We break down the pricing, model support, and when it's worth it.Read →

Frequently asked questions

Are OpenAI Dots available in the EEA?

Not in Pro plan, because the first phase excludes the EEA, Switzerland, and the UK. Yes in Business Premium, because that plan covers all regions supported by ChatGPT.

How much does an additional Dot cost?

OpenAI hasn’t announced pricing. It only says that in the future you’ll be able to purchase additional Dots and increase their speed or scope of work.

What model does Dot run on?

GPT-6 Astra, OpenAI’s most powerful model available to users.

Can Dot make transfers on its own?

No. You must always make transfers between accounts and password changes yourself. Purchases with a card saved in a store require your approval.

Do conversations with Dot use up ChatGPT limits?

Conversations don’t. Tasks that Dot assigns in Codex or ChatGPT Work count toward limits as usual.

// read moreOpenAI DevDay 2026: Dots, GPT-6.1 Sol, Pro 500 and MoreNewsGPT-6.1 Astra Halted: OpenAI Won’t Release Model Over SafetyNewsOpenAI Private Intelligence: ZDR, PSP & GDPR for EnterprisesNews

Kinetyra