OpenAI introduced dots at its DevDay conference on September 29, 2026: always-on agents inside ChatGPT that keep working between your conversations. A dot has a name, its own cloud computer and browser, memory, and the apps you connect to it. You give it a goal and boundaries; it works through the steps, brings results back for review, and asks when a decision needs you.
Dots join a field that already includes Meta’s Muse, Grok Bot, and ChatGPT’s own agent mode. What is specific here is the packaging: a persistent agent included in existing ChatGPT plans. Below: what a dot does, exactly who is eligible, how usage is counted, how stopping and deleting work, and what OpenAI has not published. We have not tested dots; the rollout is gradual and eligible accounts may not see them immediately.
ChatGPT dots at a glance
- Announced: September 29, 2026, at OpenAI DevDay; rolling out gradually
- Model: GPT-6 Astra
- Runs on: its own cloud computer and browser; optional access to one personal computer, off by default
- Reach it in: ChatGPT desktop and web, the mobile app when the supporting update is available, a voice call you start, Slack, and Microsoft Teams where available
- Set up on: the desktop app or a desktop browser only; mobile web is not supported
- Texting: coming soon
What a dot does
It holds an ongoing responsibility. OpenAI’s examples are standing jobs: keeping a sales proposal current as requirements change, preparing clips and show notes from each new interview, updating a study as results arrive, investigating reported bugs and preparing fixes for review. The dot decides when to pause and when to wake up, can run background agents in parallel, and can start Work or Codex tasks, including on a connected computer. You can open its cloud computer, take over the browser for a step, and hand control back. Our explainer on what an AI agent is covers the underlying pattern.
It researches in the background, read-only. When you are not talking to it, a dot reviews the apps you connected to find ways to help and keeps private notes. The tools used for that research cannot send messages, change app content, or control a browser or computer; any follow-up action needs the applicable permissions.
Actions go through review. Before a dot takes an action that could affect your accounts or share information, an automatic review checks it against your instructions, permissions, custom rules, and built-in safety requirements, and decides whether it can proceed, needs approval, or must be done by you. Changing a password is always yours. OpenAI’s documentation is specific about scope: asking a dot to draft replies does not give it permission to send them.
Custom rules are optional. Four behaviors can be set for a described action: take action without asking, take action when you say so, ask before taking action, or hand off to you. OpenAI describes rules as instructions the dot tries to follow and says it can make mistakes; rules do not grant app access or override built-in safety requirements.
Channels are separate from access. Connecting Slack or Teams lets you talk to the dot there; it does not connect your inbox, other apps, or computer. In Slack the dot responds to you by default, and adding it to a channel does not start monitoring until you tell it what to watch. You can call your dot; calls initiated by the dot are planned for later. It cannot have its own email address at launch.
OpenAI also previewed specialist dots with their own identity and credentials for defined roles inside companies, run as enterprise pilots and to be managed through Microsoft Agent 365. No timeline or price.
Who gets a dot
| Plan | Access | Conditions |
|---|---|---|
| Pro 100, Pro 200, Pro 500 | First dot included | Users over 18, outside the European Economic Area, United Kingdom, and Switzerland |
| Business Premium | First dot included | Rolling out worldwide |
| Enterprise | Beta, rolling out worldwide | Off by default; a workspace administrator must enable it. No data residency or inference residency during the beta. Unavailable for FedRAMP workspaces, workspaces with EKM, and workspaces with UAE inference residency |
| Plus, Go, Free, Business Standard | Not included | OpenAI says it plans to expand dots to more users |
Source: ChatGPT Learn, Meet dots and the Enterprise local-access guide, September 30, 2026. OpenAI’s announcement adds that Edu and Healthcare workspaces can try the Enterprise beta when an admin enables it; the access table in the documentation names Enterprise only. Enterprise admins get separate switches for Slack and Teams, local computer access, and custom rules, plus cloud browser, network, and computer-use controls shared with ChatGPT Work. Enterprise model controls do not apply to dots.
How usage is counted
Conversations with a dot do not count toward ChatGPT chat limits. Work and Codex tasks it starts use those products’ limits as usual. A plan also includes an allowance for the dot’s deeper work, with higher limits during the first month; OpenAI has not published the allowance size. Additional dots, and more speed or monthly capacity per dot, are described as future options without prices.
That is three different meters, and only the first is unlimited. A dot that spends its days starting Codex tasks will consume your Codex allowance whatever the dot itself costs.
Stopping and deleting
- Pause stops the dot’s current main task. It does not stop every delegated task or cancel scheduled runs; stop those in Activity and Scheduled.
- Delete removes the dot. OpenAI tells users to read the confirmation for how it handles conversations, memories, and scheduled tasks, and to save results first; it cannot be undone. The help center article calls this action Reset.
- What deletion does not do: it does not undo changes already made in connected apps or recall messages already delivered to other people.
- Disconnecting an app stops future access but does not delete information the dot already obtained from it.
- Training: Business, Enterprise, and Edu content is not used to improve models by default; personal plans follow your data controls. OpenAI says it does not train directly on proactive research or a dot’s private notes.
Should you set up a dot?
- Pro or Business Premium, eligible: start with responsibilities you can inspect and that end in a draft, such as a daily summary of calendar and open threads, a watched feedback channel with proposed replies, or a document kept current as inputs change. Set “ask before taking action” for anything that sends or deletes.
- Pro in the EEA, UK, or Switzerland, or under 18: not available; no date given.
- Enterprise: an owner must enable the beta, and the residency limits above may rule it out for regulated workspaces.
- Plus or Free: nothing to set up yet.
What is not yet known
| Question | Status on September 30, 2026 |
|---|---|
| Size of the deeper-work allowance, in month one and after | Not published |
| Price of additional dots, extra speed, or extra capacity | Not published |
| When texting launches and for whom | “Coming soon”; the help center describes a limited beta for some US Pro users |
| Teams availability | Listed as a channel “where available”; setup and admin permission required |
| Cloud computer specifications and session limits | Not published |
| Independent testing of reliability or safety | None yet; OpenAI points to a safety post and system card |