What is OpenAI dots? The always-on agents explained

The short version: dots are OpenAI's always-on agents. You create a dot in ChatGPT, give it access to the apps you choose, and it keeps working on your behalf between conversations — on its own cloud computer, around the clock — instead of waiting for your next message.

That is OpenAI's own framing: dots are capable, always-on agents designed to handle a wide range of work, powered by GPT-6 Astra, announced at DevDay on 2026-09-29 (launch post).

If you only remember one thing, remember this: a dot is not a chat persona and not a saved prompt. It is a long-running worker with its own machine, its own identity in the apps you connect, and its own set of rules about when it may act without asking.

What a dot actually is

Three properties separate a dot from a normal ChatGPT conversation.

It has its own computer. Every dot runs in its own cloud workspace with a browser, where it can read pages, analyse files, run tools, and write and test code. OpenAI says you can open that computer at any time to inspect the work. Connecting your own device is optional and separate.

It keeps working between conversations. A dot is designed to take on a project and keep making progress — including while it is handling several other projects — so you are not re-explaining context in every thread. OpenAI describes scheduled runs and proactive background updates appearing in the conversation.

It acts, within rules you set. A dot can work across the apps you connect. Before actions such as sending an email or changing a file, a separate system called Auto-review checks the planned step against your instructions, your custom rules, and OpenAI's safety requirements. Some actions always require your confirmation, and some are always handed back to you regardless of what you approve.

What dots can do

OpenAI's launch post groups the capability claims around work that normally stalls for lack of time:

• Software work. A dot watches recurring customer feedback, scopes small fixes, completes and tests them, and hands you a pull request with a video of the change.
• Release coordination. When product scope changes, it works out what that means for the launch narrative, revises the materials, and prepares drafts against the direction you set.
• Research maintenance. When new data arrives, it re-runs an analysis, investigates surprising results, updates figures, and flags what needs your review.
• Sales and technical evaluation. It checks a customer's requirements against product documentation and history, identifies untested areas, and recommends an engineer who has handled a similar deal.
• Content production. From a new interview transcript it picks clips, prepares show notes, drafts social posts, and propagates your edits across the material.

Where you can reach a dot

OpenAI says a dot is reachable through ChatGPT on web, mobile, and desktop, and also through SMS, Slack, and Teams — with context carried across channels, so starting a task in ChatGPT and adding a detail by text does not reset it. Voice calls with a dot are supported.

There are real limits at launch, and they matter more than the marketing:

Channel	Status at launch
ChatGPT desktop app / desktop web	Required to create a dot
ChatGPT mobile app	Can message your dot once mobile access is available
Mobile web	Not supported
SMS / iMessage / RCS	Limited beta, US Pro users only, not in Business or Enterprise workspaces
Slack, Teams	Announced as supported surfaces

Availability, in OpenAI's own words

This is the part most coverage gets wrong, so here is the precise version from OpenAI's help center:

• Pro: rolling out in markets excluding the European Economic Area, Switzerland, and the UK.
• Business Premium: available across all supported ChatGPT regions.
• Enterprise (and Edu, Healthcare): beta, initially off by default, enabled by a workspace admin.
• Not available to users under 18.
• For the first month, dots usage does not count toward eligible Pro, Business, and Enterprise plan allowances. After that, OpenAI says it will publish usage terms per plan.

What OpenAI has not said

We list open questions rather than guess at them:

Where dots are going

Two directions are explicit in OpenAI's launch material. Multiple dots working as a team is described as a future capability. And dedicated dots — agents with their own identity, credentials, and a defined job inside an organisation — are in a limited enterprise preview, with OpenAI trialling them internally in areas including procurement, invoice processing, email marketing, customer support, and commercial contracting. OpenAI also says it is working with Microsoft to integrate dedicated dots with the governance and security controls in Agent 365.

The honest caveat

OpenAI states in the same documentation that dots can still make mistakes, and that you should review consequential work. Auto-review, sandboxing, and monitoring reduce risk; OpenAI explicitly says they do not eliminate it. Our limitations and risks page collects those constraints in one place, because they are the difference between evaluating dots and believing a launch video.

Next: what dots cost and what counts against your limits, or set one up step by step.

Found an error or a fact that has changed since our last update? Tell us — corrections are published, not quietly edited.