OpenAI Dots Explained: What the Always-On Agents Do

ChatGPT answers when you ask. A dot keeps working after you stop asking. That is the whole pitch behind OpenAI Dots, the product OpenAI put at the centre of its DevDay keynote on 29 September 2026, and the reason the launch came with a new safety post, a help-centre FAQ and an appendix to the GPT-6 Astra system card. This guide covers what a dot is, what it runs on, who gets one, and the limits OpenAI itself lists.

What OpenAI Dots are, in OpenAI's words

OpenAI's announcement defines them in one line: "Dots are remarkably capable, always-on agents built to handle everything." Strip the marketing and three things remain. A dot is persistent, so it holds a goal across days rather than across one chat. It has its own tools, so it acts in your apps instead of describing what you should do. And it is proactive, so it looks for work between your messages.

The model underneath is GPT-6 Astra, the agent-tuned sibling of the GPT-6 models that reached ChatGPT this month (our guide to whether GPT-6 is free covers that split). OpenAI says dots are "built on existing model and agent capabilities", which is an honest way of saying the pieces, Codex, the browser agent and ChatGPT's connectors, already existed. Dots bundle them into a single named assistant that is on all the time.

How a dot works

Each dot gets "its own cloud computer", in OpenAI's phrasing, with its own browser. It signs into the apps you connect, reads and writes in them, and can run code. Through a plugin ecosystem it reaches more than 4,000 apps, per the announcement. Nothing runs on your laptop unless you grant that separately, and that permission is off by default.

When you are not talking to it, the dot runs what OpenAI calls proactive research. The safety post is specific about the boundary: background tasks use read-only tools, cannot send messages to anyone, cannot change content in connected apps, and cannot control a browser or desktop. Anything the dot wants to act on from that research has to go through the normal approval path later.

A dot also keeps notes to itself and, OpenAI says, learns your preferences and standards from feedback over time. That memory is the part with the fewest controls, and we come back to it below.

Where you talk to it

Dots live inside ChatGPT on desktop, web and mobile, including voice calls. They also sit in Slack and Microsoft Teams, so a bug report posted in a channel can start a dot investigating without anyone opening ChatGPT. Texting a dot by SMS is in a limited beta for US Pro users. OpenAI says a dot "will have the full context wherever you reach it", so a job started by voice can be checked in Slack.

Who gets a dot, and what it costs

Plan	Dot access	Notes
Free, Go, Plus	No	The pricing page marks dots unavailable on all three
Pro	First dot included	Excludes the EEA, Switzerland and the UK at launch
Business Premium	First dot included	All supported ChatGPT regions
Enterprise, Edu, Healthcare	Beta	Off by default; a workspace admin must enable it

The first dot comes "at no extra cost" with Pro or Business Premium. Pro is the catch: OpenAI now sells it in three usage tiers, and NBC News reported the range as $100 to $500 a month, with Sam Altman calling dots a premium product. Rollout is gradual and the help centre says access can take several days to reach an eligible account. Users under 18 cannot create a dot.

Two usage rules matter. Conversations with your dot do not count toward ChatGPT usage limits. Tasks the dot starts or manages inside Codex or ChatGPT Work do count, as usual, with extended limits for the first month after launch. OpenAI has said more dots and higher output tiers will be sold later but has published no prices for them.

The controls OpenAI built in

Four features decide how much a dot does on its own:

• Custom Rules. You allow, require approval for, or block classes of action, such as never sending email. The dot needs your approval to change a rule, and no rule can switch off the mandatory safeguards.
• Approvals. Permanent deletion, installing software from unrecognised sources, granting security-sensitive access and purchases on a saved card wait for you. Changing a password or moving money between financial accounts is handed back to you entirely.
• Activity View. A live log of what the dot is doing, including background work, where you can redirect or stop it.
• Auto-review. A separate system checks planned actions against your instructions, your rules and OpenAI's safety requirements before they run, and can block a step and explain why.

Credentials get their own path. For supported sign-in flows the model pauses while you type a password directly into the browser, and saved passwords come from an encrypted credential service the model never reads.

What a dot cannot do yet

OpenAI's own line is "Dots can still make mistakes, so always review consequential work." The published limits back that up:

• Memory cannot be edited. The help centre states you "currently cannot view, delete or directly modify individual dot memories." Clearing them means deleting the whole dot. Disconnecting an app stops new access but does not remove what the dot already learned from it.
• Humans may look. OpenAI says human review of a dot's activity "may occur in limited circumstances, including safety-related cases", even with model training switched off.
• Scope drifts on long chains. The dots appendix of the GPT-6 Astra system card, as reported by The Rundown, found flags for moderate scope violations rising from 8.6% with five intervening tasks to 19.7% with ten, with no severe breach or data loss in the test.
• The web still blocks bots. A two-day hands-on test reported on 7 October had a dot refine couch recommendations and watch product pages for sales, then fail to cancel a subscription because a puzzle CAPTCHA stopped it.
• Stage demos stumbled. At DevDay the voice demo failed to start and a dot went silent for roughly ten seconds, per NBC News.

How to set one up

1. Open the ChatGPT desktop app or ChatGPT in a desktop browser on an eligible plan. Dots cannot be created on mobile.
2. Connect the apps you want it to use. Fewer is safer to begin with.
3. Name the dot and let it introduce itself.
4. Set Custom Rules before giving it a real job, especially for email and purchases.
5. Message it from the mobile app, Slack or Teams once it exists.

If your plan does not include a dot, Claude runs scheduled, unattended tasks on its Pro plan, and Microsoft Copilot is where OpenAI's enterprise specialist dots will be governed through Agent 365. Our ChatGPT vs Claude comparison covers the assistants themselves, and we have separate guides on what OpenAI Dots can do and the OpenAI Dots competitors.

FAQ

Is OpenAI Dots free?

No. The first dot is included with ChatGPT Pro or Business Premium at no extra charge, and Enterprise workspaces can enable a beta. Free, Go and Plus do not include a dot, and OpenAI has not announced a cheaper tier.

Is OpenAI Dots available in India?

Pro access excludes only the European Economic Area, Switzerland and the UK at launch, so Indian Pro and Business Premium subscribers are eligible, per OpenAI's release notes. Rollout is gradual and can take several days to appear in an account.

What model powers OpenAI Dots?

GPT-6 Astra. OpenAI withheld the newer GPT-6.1 Astra, with its safety lead telling NBC News it "didn't quite meet the bar in terms of staying within scope and authorization."

Does a dot count against my ChatGPT usage limits?

Chatting with your dot does not. Work it runs through Codex or ChatGPT Work counts against those allowances as normal, with extended limits in the first month.

Can I delete what my dot remembers?

Only by deleting the dot. OpenAI's help centre says individual memories cannot be viewed, corrected or removed, and that turning off ChatGPT Memory or disconnecting an app does not erase what the dot already received.