OpenAI has rolled out dots, a new feature that allows ChatGPT to keep working after a conversation has ended. A user can assign a job to a dot, set its access privileges and return to review the work. Each dot has its own cloud computer and browser, and the feature runs on GPT-6 Astra.
Dots were announced by openAI on September 29, 2026. The main difference to a normal ChatGPT conversation is that the work can continue after the user left. Instead of going back to the same task over and over, a dot can continue to work within the boundaries the user has created.
What is a dot from OpenAI?
OpenAI describes a dot as an agent that can keep working in between conversations. The user sets a goal and decides which apps or information it is allowed to access. The dot can do the assignment, come back when there’s something to review, or when a decision needs a human.
Each dot is a complete cloud computer. By default, access to the user’s own computer is off. If someone wants to use a dot with local files or use a browser on that machine, access must be given via the chatgpt desktop app.
This is handy if you can not finish the work in one sitting. New data can lead to updates to a research project. Sales proposal can be changed as requirements change. A product team can drop a dot in the feedback from customers and come back to the fixes reviewed later.
Work done by GPT-6 Astra
Dots run GPT-6 Astra. A dot can work with a browser, connected tools, scheduled jobs and longer assignments, OpenAI says. It can also delegate parts of a job to subagents as needed.
A normal ChatGPT discussion is awaiting the next message. A dot can continue with an approved task without the user having to restart the work each time.
Dots can also contain information relevant to their assignment. They may use relevant ChatGPT memories and information from apps approved by the user. This can be useful for projects that take several days, as instructions and corrections do not have to be repeated from the beginning each time.
Dots can connect to over 4,000 apps
OpenAI says its plugin system connects dots to more than 4,000 apps. This could be information from email, calendars, Slack, documents and business software.
The user still decides what gets connected. If access is approved, the dot can use that information as part of the job, without having to copy everything by hand into ChatGPT.
People can chat with a dot using ChatGPT and Slack, and support for Microsoft Teams is also being introduced, OpenAI says. We have voice and text messaging began with a limited beta for eligible Pro users in the US.
Dots can also be programd
You may set a dot to work later, or ask it to repeat a job at intervals. It could be asked to review a calendar in the morning, check in on new information, or go back to an unfinished assignment later in the day.
“OpenAI does proactive research, too.” This mode would let a dot read information from connected apps when the user isn’t around. That’s a read-only background, so the dot can’t send messages, edit content or control a computer in that research session, says OpenAI.
Users control the actions of a dot
Users can decide which apps a dot can access, and which actions it can take. Custom Rules can allow the action, block it, request approval, or send the decision back to the user.
Certain actions are tied to the person using the account. A simple example is changing a password at OpenAI. Users can also see what a dot is doing and redirect the work if needed in Activity View.
OpenAI advises users to still fact-check important work, as dots can get things wrong. The feature is intended to move work forward within constraints and not make every decision on its own.
OpenAI is also experimenting with niche dots
OpenAI is testing company work specialist dots. They can have their own company identity, credentials and access rights to a particular job.
“Early work has included buying, invoice processing, email marketing, customer support and commercial contracts,” says OpenAI. Microsoft is also collaborating with OpenAI to introduce specialist dots into Microsoft Agent 365, allowing businesses to manage them using existing access and security controls.
Who can use dots?
Sep 29, 2026: Dots rollout started by OpenAI. Access is being deployed gradually to ChatGPT Pro users, and the European Economic Area, Switzerland, and the United Kingdom are not included in the first Pro rollout.
Business Premium users can access dots in supported areas to use with the chat GTP. Enterprise customers can test the beta if an admin enables it. OpenAI states that Enterprise access is off by default.
The first dot must be made via ChatGPT on desktop web or the ChatGPT desktop app. Once set up, it can also be accessed from the ChatGPT mobile app where supported.
Now Open a Job can hold a job open
The basic idea behind dots is very simple. You can give a task to the chatbot and walk away without stopping the work. The dot can proceed with the approved assignment, gain new information, come back when something needs to be reviewed, and keep the same work in an active state over time. That is what separates dots from the regular ChatGPT formula. The conversation might be done, but the job can go on.