An OpenAI Dot is an always-on agent in ChatGPT, powered by GPT-6 Astra, that keeps working toward goals you assign and asks you when a decision needs your judgment. As of September 29, 2026, it lives on a cloud computer OpenAI hosts, talks to you in ChatGPT (and, if you connect them, Slack or Microsoft Teams), and can use plugins you already allow.

OpenAI’s ChatGPT documentation puts it in one line: your Dot is an agent that keeps work moving across tools and projects. You keep talking to it while it works. It reaches out with results or with a decision that needs you.

#What is a Dot, in practical terms?

A Dot is not a new chat you open for one prompt. You create one primary Dot, give it a name, and treat it as a standing colleague with a handle (the default is @yourname-dot; naming it changes the handle, for example @tibo-alfred for a Dot named Alfred).

OpenAI’s launch post says Dots have their own cloud computer, learn from feedback, and can work toward your goals 24/7. Through plugins, they “readily connect to over 4,000 apps.” You can open the Dot’s computer to inspect what it is doing. You can also permit it to use your laptop.

ChatGPT documentation adds the mechanics:

  • The Dot can research, analyze data, prepare documents, and build software, using context from past conversations and your preferences.
  • It can run several pieces of work in parallel through background agents while you keep talking to it.
  • It can start ChatGPT Work or Codex tasks, including coding work in a Codex cloud environment you already created.
  • Switching from ChatGPT to Slack, Teams, or a voice call does not create a second Dot or wipe its saved notes.

If you want the setup clicks, use set up your first Dot. If you want plan and country rules, use how to get OpenAI Dots.

#How is a Dot different from ChatGPT, Work, and Codex?

OpenAI still sells those products. A Dot sits on top of them.

A normal ChatGPT conversation waits for you. A Dot takes a responsibility (“keep the offsite on track,” “watch checkout bugs”) and continues after you close the window. ChatGPT Work and Codex remain the places heavier document and coding tasks run. The launch post is explicit: talking to your Dot does not count toward ChatGPT usage limits; Work and Codex tasks the Dot starts still count as usual.

OpenAI’s launch post describes a developer Dot watching customer feedback, preparing and testing fixes, and bringing pull requests for review. That is the Dot as dispatcher, Codex as the coding runtime.

OpenAI’s safety post says you give Dots goals and define what they can do on their own. They adapt as things change and keep making progress between conversations. They turn to you for decisions that need judgment.

For a sourced comparison with older agent products, see Dots vs ChatGPT agent vs Codex.

#What can a Dot actually do?

OpenAI’s launch page walks through five jobs it has already shown:

Job OpenAI describes

What the Dot does in that example

Turn feedback into tested fixes

Watch customer feedback, scope smaller fixes, build and test them, bring complete pull requests with videos

Revise a launch as scope changes

Learn audience and standards, rewrite launch materials when the product changes

Update analysis with new evidence

Rerun analyses, investigate surprises, update figures and the supporting explanation

Update a proposal as requirements shift

Check requirements against docs, build a proof of concept, flag open concerns

Automate content production

Find clip-worthy moments in a transcript, draft show notes and social posts for approval

ChatGPT documentation gives smaller, copy-paste starters: keep an offsite checklist current, prepare captions from an interview, flag product changes that affect launch copy, update a study chart as measurements arrive, investigate a missing sidebar item and prepare a fix.

Two limits sit next to those demos. OpenAI says Dots can still make mistakes, so you should review consequential work. Proactive research (background reading of connected apps) is read-only: it cannot send messages, change app content, or drive a browser or computer. Follow-up actions go through the same permissions and Auto-review as anything you asked for.

#How does the cloud computer work?

Each Dot has a cloud computer and a browser. Work there can continue while your devices are off. The computer keeps files, software, and its own website sessions. Signing into a site on your laptop does not sign the Dot in.

When a site needs a login, the Dot can send a private sign-in form. Credentials go to the remote browser, outside the conversation. OpenAI’s safety post describes the same flow: the model is paused while you complete the form. Saved-password reuse, when offered, still needs your confirmation.

You can take over the cloud computer’s mouse and keyboard, then return control. If a site blocks the cloud browser, you can ask the Dot to try a connected personal computer. That may start a separate local task. It does not copy the cloud browser’s cookies over.

Connecting your own computer is optional and starts off. Documentation says you connect one personal computer at a time from the ChatGPT desktop app, keep it online with the app open, and revoke access when you want the Dot off that machine. Local skills need that connection. Messaging the Dot in Slack does not grant laptop access.

#How does memory work across Slack and Teams?

The Dot uses the current conversation, relevant ChatGPT memory, and its own saved notes about preferences, decisions, and ongoing work. Those notes are not a full transcript.

ChatGPT, Slack, and Teams are ways to reach the same Dot. A ChatGPT message is not automatically copied into Slack. The Dot can use a private deadline to help you prepare an update, but posting that deadline into a team channel still needs permission. OpenAI’s launch post says Dots carry context across channels so you do not re-explain the project.

The Enterprise admin guide limits Teams access to an invite-only alpha. Channel setup is covered in Dots in Slack and Microsoft Teams.

#What model is behind Dots?

GPT-6 Astra. OpenAI’s model page, retrieved September 29, 2026, describes Astra as state of the art on computer use, browsing, software engineering, cybersecurity, science, and professional work. Dots inherit that computer-use stack: filling forms, working in apps that have no API, writing code, drafting documents.

The same page lists API list prices of $10 per million input tokens and $50 per million output tokens for GPT-6 Astra. Dots themselves are a ChatGPT product with a plan allowance, not a metered API SKU you buy separately. More on the model is in GPT-6 Astra, the model behind Dots.

#Who can get a Dot, as of September 29, 2026?

OpenAI is rolling the personal Dot out to Pro and Business Premium in eligible markets, with the first Dot included at no extra cost. ChatGPT documentation lists Pro 100, Pro 200, and Pro 500 for users aged 18 and older outside the EEA, UK, and Switzerland, and says Business Premium and Enterprise are rolling out worldwide. Enterprise (and, on the launch post, Edu and Healthcare) can try a beta when an admin enables it; it starts off.

You create the Dot in the desktop app or a desktop browser. You cannot currently create one on mobile. Mobile web is not supported.

OpenAI also previewed specialist Dots for organizations: a company sets up each one with its own identity, credentials, and access to systems of record. Early internal tests OpenAI names include procurement, invoice processing, email marketing, customer support, and commercial contracting. Those are pilots, not the consumer primary Dot.

#What stays under your control?

You choose which apps the Dot can use, through existing ChatGPT plugin permissions. Custom Rules can allow an action, require approval, or hand a step back to you. They cannot turn off built-in safety requirements. Auto-review checks actions that could affect accounts or share information. Changing a password, permanently deleting data, and transferring money between financial accounts are examples OpenAI says stay with you or require confirmation.

On Business, Enterprise, and Edu workspaces, OpenAI says it does not use content to train models by default. On personal plans, the Improve the model for everyone setting applies to eligible Dot conversations and work. OpenAI says it does not train directly on proactive-research threads or the Dot’s private notes; if a note later appears in an eligible conversation, that setting applies.

The longer control list lives in privacy, security, and admin controls.

#What to watch next

OpenAI says it plans more Dots per account, ways to scale each Dot’s speed or monthly work, texting, and a wider rollout. As of September 29, 2026, ChatGPT Learn’s messaging page still says “Texting is coming soon,” while the live Help Center article checked the same day describes a limited Pro-only US texting beta. Treat texting as in flux until those pages agree. The personal Dot rolling out to eligible accounts is the one documented above.