OpenAI Dots are agents that live inside ChatGPT and keep working after you close the tab. OpenAI announced them at DevDay on September 29, 2026, and each one runs on GPT-6 Astra with its own cloud computer and browser. Most chatbots wait for your next message. A dot does not.
The feature is only a couple of weeks old, so most of what follows comes from OpenAI’s own announcement and help pages, as relayed by DataCamp’s breakdown and a consultancy’s enterprise review. DataCamp’s author found no independent third-party test of dots on launch day. Treat the capabilities below as claims until someone else measures them.
What OpenAI Dots Actually Do
A dot is a persistent agent you create once and then message like a coworker. It can reach more than 4,000 apps through ChatGPT’s plugin ecosystem, work on several projects in parallel, and run scheduled jobs such as a morning calendar review. When you are not talking to it, it looks for ways to help using your connected apps, but only with read-only access.
You can reach a dot from ChatGPT on desktop, web and mobile, and from Slack and Microsoft Teams. Setup is desktop-only. Texting is a limited beta for US Pro users. OpenAI’s own examples include a developer dot that watches customer feedback and returns pull requests with videos of the change, and a research dot that reruns an analysis when new data arrives.
This is the same shift we covered in our guide to AI agents in 2026: the assistant stops being a window you type into and becomes something with a job. Google took a similar route with the Gemini agent and its own email address.
How Much Control You Get
The safety design matters more here than the feature list, because a dot can act while you are away. OpenAI built in four layers:
- Custom Rules: each rule is one of four behaviors: act without asking, act if pre-approved, ask first, or hand the task to you.
- Auto-review: actions that affect your accounts or share information are checked against your instructions and OpenAI’s safety requirements before they run.
- Fixed hand-offs: password changes and money transfers always come back to you, and rules cannot turn that off.
- Local access off by default: a dot only touches files on your computer if you connect the desktop app and confirm.
Dots can also sign in to supported sites with saved passwords without the model ever seeing them. OpenAI itself describes its prompt-injection protection as “not perfect,” and warns that dots can make mistakes on consequential work. The innFactory review points to the combination that worries security teams: a dot that reads email, browses the web and can send messages sits where private data, untrusted content and an outbound channel meet.
Who Can Use OpenAI Dots Right Now
Access is narrow. Your first dot is included at no extra cost on ChatGPT Pro and Business Premium. Enterprise, Edu and Healthcare workspaces get a beta that an admin must switch on, and it is off by default. Two gaps stand out:
- Pro users in the EEA, Switzerland and the UK do not get a dot at launch, and OpenAI has not said why.
- There is no public API for dots. GPT-6 Astra itself is available through the API, but the agent product is not.
Pricing after the opening period is unclear. Conversations with a dot do not count toward usage limits, but deeper background work draws on a plan allowance, and tasks it starts in Codex or ChatGPT Work count as normal. Reports differ on whether dot usage is free for the first month, and OpenAI has not published the terms that follow. Teams of dots and extra dots per user are promised for later.
The Memory and Cleanup Catches
Dots inherit your ChatGPT memories and build their own, including from connected apps. At launch you cannot view or edit a dot’s memory. To delete what it learned, you delete the dot. Disconnecting an app does not remove what the dot already took from it, and files or Codex threads a dot created are stored separately and survive a reset, so you have to check them yourself.
If you have read our piece on why agents forget things, this is the opposite problem: an agent that remembers a lot and gives you few tools to audit it.
Should You Try a Dot?
If you are on Pro or Business Premium, the first dot costs nothing extra, so a low-stakes trial is cheap. Start with read-heavy chores: a daily calendar summary, a weekly digest of customer feedback, a research task that reruns when data changes. Keep Custom Rules strict at first, leaving email, calendar changes and spending on “ask first,” and check what apps are already connected to ChatGPT, since a dot inherits them.
Be more careful with anything involving money, customer data or outbound email until independent testing exists. OpenAI also held back a planned GPT-6.1 Astra upgrade after it failed internal safety tests, according to innFactory’s review. That lines up with the concerns in our look at why OpenAI flagged its own model as critical. Dots are a real step toward agents that work in the background, and they ship with the right kinds of controls. They also ship with no independent track record yet.
Sharing Is Caring
One thought on “OpenAI Dots Explained: ChatGPT’s Always-On Agents and Their Limits”