OpenAI Starts Rolling Out Always-On ChatGPT Agents That Work Between Conversations

A ChatGPT task no longer has to end when the conversation does. OpenAI began gradually rolling out dots on September 29, 2026: always-on agents that can use connected apps, keep context and continue work between chats. The shift gives users a way to delegate ongoing work, but it also makes the agent’s access and memory worth checking.

A task can keep moving after the chat

Each dot has its own cloud computer. A user gives it a goal and defines what it may do independently; it can then work through a task and bring results back for review. OpenAI says the dot is powered by GPT-6 Astra and turns to the user for decisions that need their judgment. That is a different arrangement from asking a chatbot for one answer at a time.

The dot can run a reminder or recurring check, such as reviewing a calendar each morning. It can also research in the background and suggest help without waiting for another message. Connected apps add another source of context: the dot can review their information proactively and form memories from it, even without a specific new request.

Setup starts in the ChatGPT desktop app or on desktop web. After creation, users can talk to a dot in the mobile app when mobile access is available, but cannot create one there; mobile web is not supported. Users can inspect work marked in progress, scheduled or completed from the dot’s profile.

Permission is a continuing choice

A dot’s cloud computer does not automatically give it access to a user’s own machine. Local-computer access starts off and must be granted through the desktop app. Once enabled, the dot can access files and work on that computer from any of its messaging channels; the user can revoke access. OpenAI also lets users pause the dot when they do not want it to keep working.

For some actions involving what a dot can share, purchase or access, users can set custom rules. The available choices are:

• Take the action without asking.
• Take it if the user explicitly requested it in the prompt.
• Ask before acting, or hand the action to the user.

Those settings are not a guarantee of correct behavior: OpenAI warns that dots can make mistakes even when following rules and tells users to check important results. A reset goes further than a pause. It deletes the dot, including its conversations, saved memories and scheduled tasks.

Who gets access first

The gradual rollout covers Pro users outside the European Economic Area, Switzerland and the UK, plus Business Premium users across supported ChatGPT regions. Enterprise users can try the beta only if a workspace administrator enables it; the setting is off by default. OpenAI says access may take several days to reach an eligible account.

For the first month, dot usage will not count toward eligible Pro, Business and Enterprise plan allowances. OpenAI says it will share each plan’s usage terms afterward. That leaves a practical question for anyone considering a recurring task: how much of that ongoing work their plan will allow once the introductory period ends.

There is also a narrower texting beta for some US Pro users, using a third-party provider. It is not available in Business or Enterprise workspaces, and OpenAI cautions against sharing sensitive information by text. For now, that channel is an option for a limited group, not a standard way for every dot user to reach an agent.