OpenAI dots explained: who gets ChatGPT's always-on agents, what they can do and how to control them

OpenAI's dots are "always-on agents" inside ChatGPT. OpenAI launched them at DevDay on September 29, 2026. Unlike a chat, which waits for your next message, a dot keeps working on a goal between conversations. It follows up when something changes and asks you when a decision needs you. Each dot runs on GPT-6 Astra and has its own cloud computer and browser. This explainer covers who can get one, what it can do, the controls and privacy rules, and how dots differ from ChatGPT Work. It is based on OpenAI's announcement, its help center and product documentation, checked on October 12, 2026.

The short version

• What: a persistent agent, powered by GPT-6 Astra, with its own cloud computer and browser. It connects to apps through plugins (OpenAI says "over 4,000") and can delegate tasks to ChatGPT Work or Codex.
• Who: Pro (Pro 100, 200 and 500) for users over 18, excluding the EEA, the UK and Switzerland, and Business Premium seats. Enterprise, including Edu and Healthcare, gets a beta that is off by default until an admin enables it.
• Cost: your first dot is included with an eligible plan. Chatting with your dot doesn't count toward ChatGPT usage limits, but the Work and Codex tasks it starts do.
• Controls: read-only background research, custom rules, approval checks and an Activity view. Some steps, such as changing a password, always stay with you.
• Caveat: OpenAI says dots "can still make mistakes, so always review consequential work."

What is a dot?

OpenAI's documentation describes a dot as "an always-on agent that keeps work moving and asks for your input when a decision needs you." In practice it combines three things that ChatGPT already had separately:

1. Persistence. A dot tracks progress, decides next steps and "can pause and wake up on its own," so you don't need a fixed schedule for every follow-up.
2. Its own computer. The dot's cloud computer and browser keep working while your devices are off. You can open it to inspect what the dot is doing, click Take over to act yourself, and Return control when you're done.
3. Memory. A dot draws on the conversation, relevant ChatGPT memory, and its own saved notes about your preferences, decisions and ongoing work.

You reach the same dot in ChatGPT on desktop, web or mobile, by voice call, or in Slack and Microsoft Teams. Switching channels doesn't create a new dot or reset its memory. Texting by phone is a limited beta for Pro users in the US.

TechCrunch's launch-day coverage noted that much of this was already possible with Codex and other agent harnesses. What's new is bundling it into a single persistent agent.

What can dots do?

OpenAI's examples are ongoing responsibilities rather than single tasks:

• watching for recurring customer requests, then building and testing fixes and delivering pull requests;
• revising launch materials, proposals or test plans when the scope or requirements change;
• re-running an analysis when new data arrives and flagging findings for review;
• tracking decisions, deadlines and budget items for an offsite from shared emails, and drafting replies without sending them;
• monitoring a Slack channel at set times and messaging you when a deadline is at risk.

To do this, a dot can start cloud threads, create Work or Codex tasks on a connected computer, continue existing local Codex tasks and use a Codex cloud environment for coding. You'll find delegated tasks in the dot's Activity view, sorted into In progress, Scheduled and Completed.

There are launch limits, according to OpenAI's help center. A dot can't call you, can't have its own email address, and can only use Codex cloud environments you've already created in Codex.

Who can get a dot, and where?

Last verified: October 12, 2026

OpenAI says rollout is gradual and "may take several days," so an eligible account may not see dots right away. You create a dot in the ChatGPT desktop app (macOS or Windows) or a desktop browser. After setup, you can message it from the mobile app. Mobile web isn't supported.

Specialist dots are a separate, enterprise-only preview. These have their own identity, credentials and system access for defined jobs, such as procurement, invoice processing or customer support. OpenAI says they start with "focused enterprise pilots," and that it is working with Microsoft to manage them through Agent 365.

How do you control what a dot does?

OpenAI layers several controls:

• App access. You choose which plugins a dot can use, and permissions live in ChatGPT's existing app controls. Read permission "may not include permission to change data or send messages." Disconnecting an app doesn't delete what the dot already obtained.
• Read-only background research. OpenAI says proactive research "uses read-only tools and cannot send messages, edit app content, or control your browser or computer." Follow-up actions need the relevant permissions.
• Custom rules. Each rule sets one of four behaviors: take action without asking, take action if pre-approved (meaning you explicitly asked for it in your prompt), ask before taking action, or hand off to you. Rules "cannot override built-in safeguards" and don't grant extra app or computer access.
• Auto-review. Automatic checks decide whether an action proceeds, needs your approval or is handed to you, especially actions that could affect accounts or share information. Asking for a draft doesn't authorize sending it.
• Steps that always stay with you. OpenAI says "certain sensitive tasks, such as changing a password, always stay with you."
• Your own computer is off by default. Connecting your laptop is a separate opt-in, limited to one personal computer that must be online with the ChatGPT app open. You can revoke access from the dot's profile.
• Sign-ins. A dot uses the same private sign-in flow as ChatGPT Work, so credentials go to the cloud browser rather than the conversation. Saved logins need your confirmation before reuse.
• Stopping. Pause stops the main task. Delegated tasks have to be stopped in Activity, and recurring tasks cancelled under Scheduled. OpenAI warns that stopping "doesn't undo actions already completed."

OpenAI also says a safety monitor can pause or stop a dot if it detects a concern.

For enterprises, admins get separate switches for using dots, adding them to Slack and Teams, local computer access, custom rules, cloud browser and network use, and the password manager. Local access and custom rules are also off by default. One quirk is worth knowing: OpenAI's admin guide warns that if any role assigned to a member grants dots access, that member can still use dots even when another role disables them. Enterprise model settings don't apply to dots.

What about privacy and your data?

According to OpenAI:

• Business, Enterprise and Edu content isn't used to improve models by default.
• On personal plans, you control whether your dot's conversations and work are used for model improvement.
• OpenAI says it doesn't train directly on a dot's proactive research or its notes to itself, though that information may inform a later conversation or task, depending on your settings.
• A dot checks that you've allowed it before sharing information from a private conversation with other people. In a Slack channel, other members can see its posts, but only the owner can direct it.
• Resetting a dot deletes its conversations, saved memories and scheduled tasks. The dot's own memories are separate from ChatGPT memory, and resetting is the only way to remove them.

OpenAI's dots documentation doesn't state retention periods, and the admin guide doesn't cover compliance logs. If you need those, ask OpenAI before a rollout. The broader risk is the same as for any agent that reads email, web pages and documents: text it reads can try to give it instructions. Our explainer on prompt injection and agents reading untrusted content covers it, and the case against unattended agents explains why narrow permissions matter.

How are dots different from ChatGPT Work and the old agent mode?

OpenAI's own comparison splits ChatGPT's surfaces this way:

The old ChatGPT agent mode (launched July 2025) ran one task at a time in a virtual browser. OpenAI's help center now says it is "no longer available" and points to Work. Our agent mode explainer traces that change. Work is still task-based: you hand it a job, now or on a schedule, and it finishes. A dot sits one level up. It holds the goal, decides when to act, and hands concrete jobs to Work or Codex. That is also why the billing differs: talking to your dot is free of usage limits, but the Work and Codex tasks it creates draw on your plan as usual.

For more on what "agentic" means in practice, see our explainer.

How much do dots cost?

OpenAI hasn't announced a separate price. Your first dot is included with Pro or Business Premium. Pro starts at $100 a month, and a Business Premium seat costs $100 a month billed annually or $125 monthly. OpenAI says plans include "extended limits for deeper work during the first month after launch," and that it plans to let you add more dots and scale each one's speed or monthly work volume. It hasn't given numbers or prices. Our ChatGPT plans comparison puts dots alongside everything else each plan includes.

What's confirmed, reported and uncertain

• Confirmed (OpenAI):
  • the September 29, 2026 launch and GPT-6 Astra model
  • plan and region eligibility, and the Enterprise beta being off by default
  • the controls, admin switches, privacy defaults and usage accounting described above
• Reported: TechCrunch's observation that dots bundle capabilities already possible through Codex-style agent harnesses.
• Uncertain:
  • when dots reach Pro users in the EEA, the UK and Switzerland
  • the size of the included allowance after the first month
  • pricing for additional dots
  • data-retention periods and audit logging for dot activity
  • when specialist dots leave enterprise pilots