How OpenAI Dots Work: Cloud Computer, Plugins and Approvals

OpenAI Dots are built around a persistent agent runtime rather than a one-shot chat reply.

Own cloud computer

Each Dot receives a dedicated cloud computer and browser. The user can open that environment to inspect progress. Optional permission can allow the Dot to work with the user's own laptop as well.

Model

Dots are powered by GPT-6 Astra, OpenAI's flagship model referenced at the 29 September 2026 DevDay announcement.

Plugins and apps

Through OpenAI's plugin ecosystem a Dot can connect to a large set of apps. OpenAI states more than 4,000. Plugin permissions are shared with other ChatGPT surfaces where applicable.

Always-on behaviour

A Dot can continue assigned work between conversations. Separately, OpenAI describes proactive background activity that uses read-only tools on connected apps—explicitly not sending messages, editing content or controlling browser or computer in that mode.

Approvals and rules

Users define what the Dot may do alone, what needs approval and what is blocked. Defaults plus custom rules form the control layer. Sensitive or irreversible actions are expected to surface for human decision.

Channels

ChatGPT (text and voice), Slack and Microsoft Teams are the primary interaction channels described at launch. Text messaging was noted as a limited beta for some US Pro users.