OpenAI launches Dots, always-on agents with their own cloud computer

OpenAI introduced Dots on September 29 at DevDay 2026. Each Dot is a personal agent powered by GPT-6 Astra that can keep working after a conversation ends, send updates and ask for confirmation when a decision is needed.

Access is rolling out across ChatGPT web, mobile and desktop for Pro, Business Premium and Enterprise plans in eligible markets. Enterprise users can use Dots when enabled by their workspace administrator.

According to OpenAI, a Dot starts with context from ChatGPT memory, uses Codex and connected tools, and can keep a project moving between sessions. Users decide which apps it can access and can redirect or pause its work.

OpenAI shows examples ranging from presentation updates to API migrations and building and testing small software changes from user feedback. Dots can also appear in Slack and Microsoft Teams while carrying context from work started elsewhere.

The Verge reports that users can create only one Dot for now; OpenAI plans to support multiple agents later.

Permissions use the same plugin connections as ChatGPT, ChatGPT Work and Codex. In its security documentation, OpenAI says Custom Rules can block specific actions, such as sending email, or require approval before proceeding.

Actions that can affect accounts or share information go through Auto-review. Operations such as changing a password remain with the user.

Dots can also perform proactive research by reading authorized connected sources and saving private notes before a request is made. That research mode cannot directly send messages, change content through plugins or control a browser or computer; follow-up actions still use the normal permission rules.

OpenAI's FAQ says a Dot can receive memories and recent conversation context from ChatGPT, while Dot conversations can contribute to ChatGPT memory. Turning Memory off stops future sharing but does not automatically delete context the Dot already received.

Disconnecting a service prevents new access but does not erase information already incorporated into the Dot's context. Deleting the Dot removes its own context, while files, Codex threads and ChatGPT conversations created during its work remain in their respective locations.

For Business, Enterprise and Edu, OpenAI says data is not used for model training by default. On personal plans, the “Improve the model for everyone” setting controls whether Dot conversations and work may be used for model improvement.

In OpenAI's examples, a Dot tracks metrics and feedback to update a presentation, checks RFP requirements against email and product documentation, prepares code and tests during an API migration, and keeps monitoring the project until dependencies are gone.

These are company demos, not independent benchmarks. The technical shift is that memory, tools, a cloud computer and background work sit under one persistent agent instead of requiring a fresh conversation for each step.

Reuters reports that some live DevDay demos ran into problems, with Dots repeatedly failing to deliver requested voice updates on stage. OpenAI attributed the glitches to rolling out all the updates at once.

Reuters also notes that the launch comes amid growing scrutiny of autonomous-agent safety. OpenAI says rules, approvals and review systems are designed to reduce the chance that an agent acts outside a user's intent.

The useful question is not whether Dots can complete a polished demo, but how often it can finish long-running tasks across real apps, permissions and data without constant human intervention.

OpenAI is rolling it out to Pro, Business Premium and Enterprise users in eligible markets. Users under 18 cannot use Dots for now, according to the official documentation.

The official product page is on ChatGPT. Dots should not be confused with domains such as dot.com or dots.com, which are not the official OpenAI product site.