OpenAI Dots are always-on agents inside ChatGPT. Each dot gets its own cloud computer, connects to the apps you pick, and keeps working on a task after you close the chat. OpenAI announced them on September 29, 2026, and this guide covers who can use them, what they cost, and how to set up your first one.

What is an OpenAI Dot?

A dot is a persistent ChatGPT agent that takes on an ongoing responsibility and keeps making progress between conversations. It runs on GPT-6 Astra (OpenAI's flagship model, released in September 2026), works from its own cloud computer, and remembers context across ChatGPT, Slack, and Microsoft Teams.

That "own cloud computer" part matters. A cloud computer is a remote machine, with its own browser, that OpenAI runs for you. Your dot does its browsing, file work, and app actions there instead of on your laptop. Your personal device stays out of the loop unless you connect it on purpose.

OpenAI's examples include a dot that reads a weekday morning calendar and summarizes it, a dot that watches customer feedback and drafts fixes for a developer to review, and a dot that notices an unsent invoice in an email thread and drafts it.

How a dot is different from ChatGPT and ChatGPT Work

In regular ChatGPT, the work happens while you're in the conversation. A dot keeps going after you leave, and it can start work on its own schedule or after a watch condition you set.

If you've used ChatGPT Work, a dot is similar in that it handles multi-step work in the background. Dots and Work share the same permissions for connected apps, and Work and Codex (OpenAI's coding agent) tasks that a dot delegates count against your usage allowance.

A dot is an AI agent with a name, a memory, and a standing assignment. An agent needs software around the model to use tools and finish tasks, and with a dot, OpenAI runs that part for you.

Who can use OpenAI Dots

Dots are rolling out gradually, and OpenAI says access can take several days to reach your account after you become eligible. The help article lists three plan groups.

ChatGPT Pro

Pro users get a first dot at no extra cost, except in the European Economic Area, Switzerland, and the UK, where Pro users don't have access at launch. Texting your dot is a limited beta for Pro users in the US.

ChatGPT Business Premium

Business Premium users get a first dot at no extra cost in all supported regions.

Enterprise, Edu, and Healthcare

These workspaces get dots as a beta. A workspace admin has to turn it on, and it's off by default. Enterprise organizations can also set up specialist dots with their own identity and credentials for focused jobs like procurement, invoicing, and customer support.

The help article doesn't list Free or Plus plans, so as of this writing those plans don't appear to have dots.

Is OpenAI Dots free?

Your first dot is included with an eligible plan, so there's no separate charge for it. OpenAI says future pricing will let you add more dots and raise speed or monthly work capacity. It hasn't published those prices.

Usage works in two parts. Talking to your dot doesn't count toward your ChatGPT usage limits. Tasks it hands off to Codex or ChatGPT Work do count. OpenAI also describes extended limits for the first month after launch, and one hands-on guide notes that the post-launch limits aren't published yet. Check Settings → Usage in ChatGPT before you leave a recurring task running.

How to set up OpenAI Dots

Setup takes about 10 minutes and has to happen on a computer. You create a dot in the ChatGPT desktop app (Windows or Mac) or in ChatGPT on desktop web. The mobile app can't create one, and neither can mobile web, but you can talk to your dot from the iOS or Android app after it exists.

The steps are the same on Windows and Mac. Menu names below come from OpenAI's help article and can change as the rollout continues.

  1. Check that you're eligible. You need a Pro or Business Premium plan (or an Enterprise, Edu, or Healthcare workspace where your admin turned dots on), and you need to be in a supported region.
  2. Open ChatGPT on your computer. Use the desktop app or the website in a desktop browser.
  3. Create your dot and follow the onboarding prompts. The first screens walk you through the rest of these steps.
  4. Name it and pick an avatar. The default handle is @yourname-dot. A custom one looks like @yourname-agentname. You also choose a character or pet avatar.
  5. Connect apps. Apps connect through plugins (connections that let ChatGPT read or act inside a service like Gmail or Slack). You can manage them later under Customize → Plugins. Start with only the apps your first task needs.
  6. Connect a messaging channel, if you want one. Slack and your personal email account are available, and you set these up from desktop only. Microsoft Teams works too.
  7. Set your custom rules. Covered in the next section. Do this before the dot does any real work.
  8. Decide about your local computer. Access to your own machine is off by default and takes a confirmation on the desktop app. Skip it unless the task needs files on your laptop.
  9. Give it a first responsibility. One clear job with clear limits works better than a long list.
  10. Review the first run. Open your dot's profile menu (•••) to check its work, scheduled tasks, and settings.

OpenAI Dots custom rules

Custom rules decide what your dot can do without asking you. They're under Customize → Custom rules, and each rule gets one of four settings.

Take action without asking

The dot does it and tells you afterward. Use this for low-risk work, like filing a summary in a folder it created.

Take action if pre-approved

The dot acts only when you've already approved that kind of action. Good for repeated tasks you've watched it handle correctly.

Ask before taking action

The dot stops and waits for your yes. This is the right default for anything that sends a message or edits a shared file.

Hand off to you

The dot prepares what it can, then passes the task back to you to finish yourself.

Some actions always need your approval no matter what your rules say. OpenAI lists password changes among them, and the launch coverage adds purchases and money transfers. An auto-review step also checks account-changing actions against your instructions, your permissions, and OpenAI's safety requirements before they run.

What a dot reads in the background

Your dot can read your connected apps without being asked, then form memories from what it finds. OpenAI calls this background research, and it's read-only: the dot gathers information and saves private notes, and any action that changes something goes through your rules.

Two details are worth knowing before you connect an inbox. Disconnecting an app stops new access, but OpenAI says information already added to the dot's context stays. And you can't delete individual dot memories; deleting what a dot has saved means resetting the whole dot, which also removes its conversations and scheduled tasks. Reset is in the profile menu (•••).

Local computer or cloud computer

The cloud computer is the default, and it's always available. Your local computer is optional. If you turn it on, the dot can work with files on your machine and create tasks there, but local work needs that computer online with ChatGPT open. You can connect one personal computer at a time, based on one hands-on guide. Codex environments you created earlier also work.

A first task to try

A good first task is small, read-only, and scheduled so you can check it. Here's one you can paste into your dot:

Every weekday at 8 a.m. Pacific, read my calendar and the last day of email. Write a private summary of meetings, deadlines that changed, and anything waiting on me, with a link to each source. Don't send, edit, or delete anything. Ask me before you contact anyone.

Run it for a week, read what it produces, and tighten the rules where it guessed wrong. Then move to a task that takes an action.

Adding a dot to Slack doesn't create a recurring schedule by itself. If you want it watching a channel, say which channel, what to watch for, how often, and when to escalate. The Scheduled tab in your dot's profile shows active, paused, and completed tasks, and you can change repeat times and notifications there.

How to pause or stop a dot

Open the profile menu (•••) and choose Pause to stop the main task. A pause may not cover everything, so also check the dot's activity for delegated tasks and the Scheduled tab for recurring runs. OpenAI also says its monitoring systems can pause a dot if they detect unsafe behavior.

Safety, privacy, and limits

One hands-on guide by Flavio Copes reports red-team results from OpenAI's launch material: no successful prompt injection attacks across 19,238 simulated email attempts. Prompt injection means hidden instructions inside an email or web page that try to steer the agent. The same guide reports that scope violations (the dot touching information outside its assigned job) rose as task chains got longer: 8.6% at five tasks and 19.7% at ten. The takeaway is to keep each responsibility narrow.

On training data, ChatGPT Business, Enterprise, and Edu workspaces exclude dot content from model training by default. On personal plans you control whether dot conversations improve models, and OpenAI says background research and dot notes aren't used for training.

Known limits at launch:

  • Dots can't call you.
  • Creating a dot and setting up Slack or texting require a desktop.
  • Texting is a limited US beta for Pro.
  • OpenAI says a dot can make mistakes, including when it follows your rules, so review its work before you rely on it.

Should you set one up?

Yes, if you have a recurring job that would never be worth building an automation for. Weekly status summaries, inbox triage, and invoice drafts fit that description. The setup is short, the first dot is included, and the read-only start keeps the risk low.

Wait if the work needs exact cost tracking per task. Dots use an opaque plan allowance rather than per-task billing, and there's no public API for dots themselves. For that kind of work, an agent built on the OpenAI API or in a tool like n8n gives you the numbers. If your work involves other people's private data, learn the memory and reset behavior first, since you can't delete single memories.

If you're new to the tools underneath this, what MCPs are for non-developers explains how apps connect to AI agents.

Sources:
Getting started with your dot (OpenAI Help Center) · Introducing dots (OpenAI) · Dots are OpenAI's new personal agents (Engadget) · OpenAI just launched Dots (The New Stack) · OpenAI launches Dots (9to5Google) · OpenAI launches dots, always-on AI agents with their own cloud computers (The Next Web) · OpenAI Dots: Always-On Agents in ChatGPT, Explained (DataCamp) · A deep dive into OpenAI dots (Flavio Copes)