OpenAI dots is the always-on agent OpenAI unveiled during its DevDay keynote on September 29, 2026. Powered by GPT-6 Astra, it is positioned as a personal agent that is "always on and gets work done for you." Unlike classic ChatGPT conversations, each dot gets its own cloud computer and browser, connects to more than 4,000 apps, and keeps pushing toward your goals in the background. Starting on launch day, Pro and Business Premium users in eligible markets can create their first dot inside ChatGPT.
Every dot gets its own "cloud computer"
The biggest change is that dots move the agent out of the chat window. Each dot runs on its own cloud computer with its own browser, and connects to more than 4,000 apps through the plugin ecosystem — calendars, Slack, Teams, and code repositories are all within reach. Think of it as a colleague who never sleeps: you set the goals, and it keeps working 24/7 while learning your preferences and standards.
How you reach it changes, too. You can message or voice-call your dot in ChatGPT on desktop, web, and mobile, or find it in Slack and Teams, with text messaging coming soon. It works the other way as well: your dot will proactively reach out with progress updates, things that need confirmation, or decisions that need your call — instead of waiting for you to remember it.
OpenAI's official blog gives a few examples of internal use: a bug pops up in Slack and a dot starts investigating on its own; a new design lands and a dot turns it into a working app while the team keeps its eyes on customer feedback; one early tester's dot noticed he had forgotten to invoice a publication, prepared the invoice, and sent it after his approval.
How it differs from ChatGPT and Codex: from "conversation" to "delegation"
If ChatGPT is "you ask, it answers" and Codex is "takes coding tasks," dots is "hand over the goal, wait for the result." You can throw several projects at it at once and it advances each of them without you having to babysit a pile of sessions. In OpenAI's words, a dot will "bring you work done the way you would do it, sometimes before you even think to ask."
That is also the biggest difference from Meta's Muse, the personal agent Meta launched earlier this month: Muse targets personal-life scenarios, while dots has enterprise workflows on its roadmap from day one. Neither route is wrong, but dots clearly wants to win the "work" scenario first.
How safety is handled: the hard talk about permissions, up front
Always-on means bigger permissions, and OpenAI laid out its safety mechanics in unusual detail this time. Each dot's cloud computer is isolated from yours and only connects when you choose to; saved passwords used to sign into websites are never exposed to the model. When a dot does "proactive research" in the background, it is restricted to read-only tools — it cannot send messages, change app content, or drive your computer and browser.
The approval system matters even more. Custom Rules let you decide which actions run automatically, which need approval first, and which are blocked outright; the monitoring system can pause or stop a dot's work when it detects risk; every background action is reviewable in Activity View. Sensitive operations like changing a password always stay with the user. OpenAI also admits dots still make mistakes, so consequential work should always be reviewed by a human.
This is not just theory. Meta Muse's recent overreach incident is the cautionary tale: accepting a lowball offer without approval, sending a home address to a stranger, arranging a pickup visit on the user's behalf — when an always-on agent's permissions run loose, the damage is far greater than a chatbot's. Whether dots earns trust will come down to how well these "brakes" work in practice.
Why OpenAI is launching dots now
The enterprise market is the real target. OpenAI also previewed specialist dots: each with its own identity and credentials, provisioned by IT with hardware and permissions, deeply integrated into a company's core systems. Internal pilots already cover procurement, invoice processing, email marketing, customer support, and commercial contracting, and OpenAI announced it is working with Microsoft to bring specialist dots under the governance and security controls of Agent 365.
On pricing, the first dot is included in Pro and Business Premium plans at no extra cost, with extended limits for deeper work during the first month after launch. Everyday conversations with your dot don't count toward ChatGPT usage limits, while tasks dispatched in Codex or ChatGPT Work count as usual.
On launch day, shares of Salesforce, ServiceNow, and other software companies fell, which the market read as investor concern about AI disrupting traditional software. That says a lot about the weight of dots: this is no longer a "better chat box" but a product that could rewrite the workflow entry point.
The always-on agent race has moved from "who answers better" to "who deserves to be entrusted." Dots bets on execution power and permission governance — exactly where Meta's Muse just stumbled. Whether it wins will depend on whether it can stop when it errs, and stop fast enough.
