OpenAI introduced Dots at its DevDay conference on September 29, a new class of always-on AI agents built on the GPT-6 Astra model. Each Dot runs on its own cloud computer and browser, separate from the user’s device.

Unlike a chatbot, a Dot keeps working toward a goal after the user stops typing. Users can name a Dot, set boundaries for what it may do independently, and let it continue until it needs review or approval.

Dots can work across more than 4,000 applications through ChatGPT plugins, including Slack and Microsoft Teams. They use tools such as Codex and ChatGPT Work to research, analyse data, prepare documents and build software.

OpenAI built in controls for sensitive actions. Changing passwords or permanently deleting data requires explicit user authorisation. The company says enterprise customer data is not used to train models by default.

The rollout began with ChatGPT Pro and Business Premium subscribers in eligible markets. The launch was not without stumbles: OpenAI’s live demos failed on the official stream.