OpenAI Dots bring an always-on agent to ChatGPT, with approvals still in the loop

An early tester’s dot spotted a forgotten invoice, prepared it, and sent it only after approval, OpenAI says in its launch announcement. That is the product in miniature: work that continues without another prompt, with a person brought back in for a consequential step. Dots began rolling out in ChatGPT at DevDay on September 29, 2026. They are not a developer API to embed in an app.

A dot is an always-on agent powered by GPT-6 Astra. It has its own cloud computer and browser, can work on several projects, and keeps context across conversations and the connected apps you permit. Users start with a primary dot, give it a name, and can ask it to manage work over time rather than opening a fresh thread for every step. OpenAI says it can use plugins spanning more than 4,000 apps; that is an ecosystem claim, not a promise that every integration or action is enabled for every account. The company also previewed specialist dots with their own organizational identities and access, but those are starting as focused enterprise pilots, not generally available seats you can add today.

What it can reach

The cloud computer is separate from your laptop. Connecting a local computer is optional and off by default; a user must enable access in the desktop app and confirm it. If enabled, the dot may work on local files and use the local browser, subject to the applicable rules. That is a larger permission than letting it work only in its own cloud environment. Its plugin permissions are shared with ChatGPT, ChatGPT Work and Codex, and can be managed in the Plugins tab. Disconnecting an app stops new access through that connection; it does not erase information already incorporated into a dot’s context, as the privacy FAQ explains.

“Always-on” also has a narrower meaning than unrestricted autonomous action. A dot may carry out a task you gave it, or run reminders and recurring checks you scheduled. Its separate proactive research mode can read permitted connected sources and save private notes without a new prompt, but its research tools cannot directly send messages, edit connected app content, or operate a browser or computer. Follow-up actions still pass through the normal action rules. Users can inspect ongoing work in Activity View, steer or pause it, and set Custom Rules that allow, require approval for, or block supported actions. OpenAI’s safety description says those rules cannot disable mandatory checks.

Before actions such as sending an email or changing a file, OpenAI says a separate Auto-review system checks the planned step against instructions and safeguards. Some steps can proceed under prior, specifically scoped approval; others require a fresh confirmation. Password changes and transfers between financial accounts are handed back to the user, while purchases with cards saved at merchants require approval. These controls reduce risk rather than eliminate mistakes or prompt-injection attacks. An approval prompt should not be read as a guarantee that the proposed action is correct, and a completed action may not be reversible.

Who can use a dot, and what counts against a plan

The current Getting started guide narrows the rollout. Pro users are eligible outside the European Economic Area, Switzerland and the UK; Business Premium users are eligible across supported ChatGPT regions. Enterprise users can try the beta if a workspace admin enables it, and it starts off by default. Rollout is gradual, so an eligible account may wait several days. The Help Center also says dots are unavailable to under-18 users. Users create a dot on desktop web or in the desktop app, not on mobile; after setup, mobile-app messaging becomes available when that access reaches the account. Mobile web is not supported.

The first dot is included with Pro or Business Premium at no extra charge. OpenAI says ordinary conversations with it do not count toward ChatGPT usage limits; tasks it starts or manages in Codex or ChatGPT Work still count toward the usual limits. The launch post describes an allowance for deeper work and extended initial limits, but gives no durable numeric quota or add-on price. The Help Center says dots usage will not count toward eligible Pro, Business and Enterprise plan allowances for the next month, with plan terms to be shared afterward. That introductory treatment should not be mistaken for unlimited work indefinitely or a published price for additional dots.

Dots should also not be confused with OpenAI’s Agents API, which gives an application access to a managed Codex harness, sessions, tools and optional sandboxes, billed at API model, tool and container rates. Dots are a ChatGPT user-facing agent with connected personal or workspace context and plan-based access. For anyone evaluating one, the first practical decision is less about the Astra model than about which apps to connect, whether to enable local-computer access, and which actions must stop for approval.