ChatGPT Dots usage limits and permissions: what runs automatically?
A practical decision guide for Pro users covering Dots availability, temporary launch limits, chat versus Work/Codex usage, background research, app access, and approvals.
Contents
The practical answer: enable a dot only if you want an agent to carry ongoing responsibility between conversations—not merely another chat window. As of October 1, 2026, eligible users can try Dots during a gradual rollout, but the usage policy has three separate layers: conversations with your dot, the dot’s own deeper work, and tasks it delegates to ChatGPT Work or Codex. Those layers are not billed or limited in the same way.
The safest starting point is to connect one low-risk app, leave local-computer access off, require approval for external actions, and watch both Activity and Settings → Usage during a small pilot. OpenAI has explicitly said that post-launch usage terms will be published later, so no one should treat the launch window as a promise of permanently unmetered work.
The current rules in one table
| Question | Current official position | What it means for you |
|---|---|---|
| Who can get a dot? | Pro 100, Pro 200 and Pro 500 users over 18 outside the EEA, UK and Switzerland; Business Premium is rolling out worldwide; Enterprise is off by default and needs an admin. | An eligible plan does not guarantee immediate access because rollout is gradual. |
| Does chatting with a dot use ChatGPT limits? | No. OpenAI’s Dots guide says conversations with your dot do not count toward ChatGPT usage limits. | Brainstorming, clarifying instructions and reviewing progress are a different meter from delegated work. |
| What about the dot’s own deeper work? | OpenAI says plans include an allowance, with extended launch limits. The latest release notes add that Dots usage will not count toward eligible plan allowances for the next month, after which plan terms will be shared. | Treat this as a temporary launch policy, not a permanent free tier. |
| What if the dot starts Work or Codex tasks? | Tasks the dot starts or manages in Work or Codex count toward those products’ usage limits as usual. | A long coding or document task can still consume the same allowance you would use by starting that task yourself. |
| Can it act in connected apps while you are away? | Assigned or scheduled work can use actions you authorized. Proactive research is more restricted and read-only. | “Works in the background” does not mean “may freely send, edit, buy or delete.” |
| Can it use your laptop? | The dot has its own cloud computer. Connecting your computer is optional and separate. | Start without local access unless the task truly requires local files, apps or a browser session. |
Which plans include Dots—and what is still unknown?
OpenAI’s current Meet dots documentation lists all three Pro tiers—Pro 100, Pro 200 and Pro 500—for adults in launch-supported markets outside the European Economic Area, the United Kingdom and Switzerland. Business Premium is rolling out worldwide. Enterprise access is also rolling out worldwide, but a workspace administrator must enable it because it starts off by default.
Free, Go and Plus are not listed as launch-access plans. Even on an eligible plan, the feature may take time to appear. Create the dot from the desktop app or a desktop browser; mobile access may follow after setup when the supporting app update is available, while mobile web is not supported.
Two cost points matter:
- You still need an eligible underlying subscription. “Included” does not mean the qualifying plan itself is free.
- OpenAI has not published the permanent post-launch Dots usage terms. The Help Center says it will share terms for each plan after the temporary window. Prices for additional dots, higher speed or more monthly work are also not public in the sources reviewed for this article.
Therefore, a plan comparison that claims a fixed number of tasks, a permanent zero-cost allowance, or a predictable cost per assignment would be speculation.
How usage is counted: three buckets, not one
1. Conversation with your dot
Talking to your dot, refining a goal, asking what it is doing, or reviewing its explanation does not count toward normal ChatGPT usage limits, according to the official Dots guide. This is the conversational control layer.
That does not mean every activity mentioned in the conversation is free. A sentence such as “run the test suite, revise the spreadsheet and prepare the deck” can cause the dot to create separate Work or Codex tasks.
2. The dot’s own deeper work
OpenAI says the plan includes an allowance for deeper work and describes extended limits during the first month. Its current release notes use even broader temporary wording: Dots usage will not count toward eligible Pro, Business and Enterprise plan allowances for the next month, and terms will be announced afterward.
The responsible reading is: the launch benefit is temporary, and the long-term unit of usage is not yet fully published. Check your account’s Usage page rather than relying on a third-party estimate.
3. Work or Codex tasks delegated by the dot
OpenAI separately states that tasks your dot starts or manages in ChatGPT Work or Codex count toward those products’ usage limits as usual. This distinction is the most important budgeting rule.
For example:
- Asking the dot to discuss a refactoring plan is a conversation.
- Asking it to create a Codex cloud task that changes a repository is delegated Codex work.
- Asking it to produce a finished spreadsheet or report through Work is delegated Work usage.
The temporary Dots launch language does not explicitly cancel Work or Codex’s own meters. Until OpenAI says otherwise for your account, budget delegated tasks as normal Work/Codex consumption.
What “proactive research” can do in the background
A dot can keep working between conversations, run scheduled tasks and look for useful updates. But OpenAI separates proactive research from assigned or scheduled work.
According to Tasks and memory, proactive research may read information from sources you have permitted and keep private notes that connect new findings to earlier work. Its research tools cannot directly:
- send messages to other people;
- change content in connected apps;
- control a browser or computer.
The dot can surface a suggestion or ask a question. Any follow-up action then goes through the normal permission and approval system.
A recurring task is different. If you explicitly schedule “check the project channel every weekday and update the checklist when a deadline changes,” that task may include actions within the scope you authorized. Connecting an app alone does not create monitoring, and proactive research alone does not grant write access.
The permission stack: access, instructions, rules and approval
It helps to think of control as four layers.
App and computer access
A dot can use only the plugins, connected accounts and environments made available to it. Your existing ChatGPT app permissions apply. Messaging it in Slack does not automatically connect your email, cloud drive or laptop. OpenAI’s computer and app guide treats those as separate connections.
Your instructions
A clear instruction defines scope: which account, which people, what action and under what condition. “Draft replies” is not the same as “send replies.” A vague goal creates more room for the agent to choose an interpretation.
Custom Rules
Custom Rules can tell the dot to act without asking, act only when explicitly requested, ask before acting, or hand the step back to you. They are useful for durable boundaries such as “never send customer messages without approval” or “hand off deletion of shared files.”
Rules do not grant access to an app, and they cannot override built-in safety requirements. OpenAI also warns that the dot can make mistakes while following rules.
Automatic review and human approval
Before an action affects an account or shares information, automatic review checks the action against your instructions, permissions, Custom Rules and safety requirements. The outcome may be: proceed, ask you to approve, or require you to take over. Changing a password is an example of a step you must perform yourself. More sensitive financial or security actions may also require takeover.
Review consequential work even when no approval prompt appears. Approval is a control, not a guarantee that the result is correct.
Cloud computer, local computer and data retention
Every dot has its own cloud computer and browser. It does not automatically inherit your laptop files, personal browser logins or device policies. You may inspect cloud work and take over the browser when needed.
Connecting your own computer is optional. Once connected, the dot may use local files, code, apps and browser access for supported tasks; the computer must remain online with the ChatGPT desktop app open. Revoke access when local work is no longer needed.
Connected apps also create a retention question. Disconnecting an app stops new access, but it does not necessarily erase information the dot already incorporated into its context. OpenAI’s privacy and safety FAQ says deleting the dot deletes its own context, while files, ChatGPT conversations and Codex threads may be stored separately and need separate cleanup.
For personal plans, review the “Improve the model for everyone” setting. Business, Enterprise and Edu data is not used for model training by default. OpenAI says it does not train directly on proactive research or the dot’s private notes, although information brought into an eligible conversation or task may follow the applicable data setting.
A low-risk way to enable Dots
Use a bounded pilot rather than connecting everything during onboarding.
- Choose one reversible job. Good examples are summarizing a read-only project folder, tracking a non-sensitive deadline or preparing a draft that cannot be sent automatically.
- Connect only one necessary app. Review both the account selected and the actions the plugin is allowed to perform.
- Leave your local computer disconnected. Add it only when cloud work cannot complete the job.
- Write explicit boundaries. State the allowed sources, output destination, notification conditions and actions that require approval.
- Use a conservative Custom Rule. Require approval before sending, purchasing, publishing, deleting or changing shared data.
- Inspect Activity and Scheduled separately. Pausing the main dot does not automatically stop every delegated task or cancel future schedules.
- Check Usage after real work. Compare the account’s usage before and after a Work or Codex task. Do not infer future pricing from the temporary launch period.
- Expand permissions one step at a time. Add write access or local-computer access only after the read-only pilot behaves as expected.
Should you enable it?
| Your situation | Recommendation |
|---|---|
| You have a multi-day project, recurring research or several related tasks that need continuity. | Yes, with a narrow pilot. Persistent context and background coordination are the main benefits. |
| You mainly ask one-off questions and do not need follow-through. | Probably not yet. Regular ChatGPT is simpler and requires fewer permissions. |
| The work involves customer messages, payments, account security or irreversible deletion. | Only with strict rules and manual takeover. Keep the highest-risk step with a person. |
| You cannot tell which Work/Codex tasks consume allowance. | Wait or test carefully. Usage visibility is necessary before scaling. |
| You must connect a highly sensitive laptop or broad company account on day one. | Do not start there. Begin with the cloud computer and least-privilege app access. |
Frequently asked questions
Does the first month remove every usage limit?
No official source says every kind of work is unmetered. OpenAI currently says Dots usage will not count toward eligible plan allowances for the next month and that future plan terms will be shared later. Work and Codex tasks are still documented as counting toward their own usage limits as usual.
Does a dot silently send emails during proactive research?
No. Proactive research tools are restricted from sending messages, changing app content or controlling a browser or computer. Sending can occur only through separately authorized work and remains subject to permissions, rules and review.
Does connecting Slack give the dot access to Gmail or my laptop?
No. Messaging channels, apps and computers are separate connections. Each needs its own authorization.
Does disconnecting an app delete everything the dot learned from it?
No. It stops new access, but existing context may remain. Deleting the dot removes the dot’s own context; separately stored files, conversations and Codex threads may require separate deletion.
Can I pause everything with one button?
Not necessarily. Pausing the dot’s current main task, stopping delegated tasks and canceling recurring schedules are separate controls. Review Activity and Scheduled independently.
Bottom line
Dots is most useful when you want continuity: an agent that remembers a responsibility, watches permitted information, coordinates work and returns for decisions. Its value is not “free, open-ended automation.” The sensible decision is to separate conversation from delegated tasks, separate read-only research from authorized actions, and separate cloud access from local access. Start with least privilege, keep consequential actions behind approval, and recheck the official usage terms after the launch window ends.
