OpenAI Dots are always-on agents. Their most important launch feature is the control boundary

[OpenAI launched Dots](https://openai.com/index/introducing‑dots/) on September 29 as always‑on agents that keep working across projects instead of waiting inside one chat. Each dot runs on its own cloud computer and browser, learns from feedback, and can connect to more than 4,000 apps through OpenAI's plugin ecosystem. GPT‑6 Astra powers the system. The durable change is continuity. A user can hand over several projects, keep adding work and inspect the dot's computer while it runs. OpenAI says dots can carry context across ChatGPT, Slack and Microsoft Teams, and users can message or call them. The launch examples span software fixes, changing launch materials, scientific analysis, enterprise proposals and content production, but these remain illustrative scenarios from OpenAI rather than independently verified completion rates. The new [Dots tool record](https://opentools.ai/tools/dots) keeps the stable product and availability facts. The [OpenAI organization profile](https://opentools.ai/organizations/openai) supplies the parent entity, while this article focuses on what the launch means for people deciding whether to delegate persistent work.

[OpenAI says Dots are rolling out](https://openai.com/index/introducing‑dots/) in eligible markets to ChatGPT Pro and Business Premium. Enterprise, Edu and Healthcare workspaces can try the beta only after an administrator enables it; the recap says it is off by default. OpenAI includes the first dot in Pro or Business Premium without an extra charge and provides an allowance for deeper work, with extended limits during the first launch month. It has not published a simple unlimited‑work entitlement. Conversations with a dot do not count against ordinary ChatGPT usage limits. Tasks the dot starts or manages in Codex or ChatGPT Work still count against the relevant usage limits. That distinction matters for budgeting: a team can talk to its dot freely while the underlying coding or professional‑work execution consumes the service allowance where the work occurs. OpenAI says users will eventually be able to add more dots and increase either speed or monthly work capacity. Those options are future‑tense. The product available today is a primary dot, with mobile messaging after setup and specialist enterprise dots limited to focused pilots.

OpenAI draws a useful boundary around work that starts without a live prompt. [Under OpenAI's proactive research rules](https://openai.com/index/introducing‑dots/), when a user is away a dot can perform “proactive research” through already connected apps, but the tools in that mode are restricted to read‑only access. OpenAI says they cannot send messages, change app content, or control a browser or computer. Action work uses a different permission path. Users choose the connected apps and can set Custom Rules that allow an action, require approval or block it. OpenAI says dots use an auto‑review system to check actions that could affect accounts or share information. Some sensitive tasks, including changing a password, always remain with the user. This distinction prevents a common reading error. “Always on” describes availability and observation, not standing permission to mutate every connected system. A rollout test should separately evaluate what the dot can observe in the background, what it can prepare for review, what it can execute under an explicit rule and what it will always hand back.

Each dot's cloud computer is separate from the user's computer unless the user deliberately connects another device. For supported websites, OpenAI says saved passwords can be used without exposing them to the model. Activity View lets users watch progress and redirect work, while monitoring can pause or stop a dot when it detects a safety concern. Data controls follow the surrounding plan. [OpenAI says](https://openai.com/index/introducing‑dots/) it does not use content from Business, Enterprise or Edu workspaces to improve models by default. Personal‑plan users control whether eligible conversations and work are used for improvement. OpenAI also says it does not train directly on proactive research or a dot's private notes, although information from that work can inform an eligible conversation or task depending on settings. A useful adoption checklist starts with one bounded responsibility, the minimum set of connected apps and an explicit approval matrix. Review read‑only background access separately from write access, inspect the first complete task in Activity View, test prompt‑injection and stale‑context cases, and define the point where the dot must stop. Persistent agents create value by remembering goals and carrying work forward; the same persistence makes permission and correction history part of the product decision. *Figure: Dots launch capabilities and control boundaries. Source: [OpenAI's Dots announcement](https://openai.com/index/introducing‑dots/), published September 29, 2026. Figure by OpenTools Team; no third‑party expressive material used.*