Complete OpenAI Dots Guide

A normal chatbot responds when you message it. A Dot can take responsibility for work over time: it can continue between conversations, monitor connected information, run recurring tasks, and surface results or decisions.

Persistent

continues work

Connected

uses approved apps

Visible

inspect activity

Bounded

asks when needed

Dots can manage ongoing tasks, remember relevant context, use connected apps, and communicate their progress—not just answer the prompt in front of them.

Astra supplies the planning and agentic reasoning behind a Dot: understanding goals, using tools, adapting to new information, and knowing when to ask.

Every Dot has a cloud computer, browser, and supported tools. It can keep working even while your personal computer is off.

Depending on availability and permissions, a Dot can work across tools such as Drive, Calendar, GitHub, Slack, Teams, Codex, Docs, Sheets, and Slides.

Dots can look for useful work in connected information. Background research is restricted to read-only access, so it cannot send, edit, or control.

Set recurring responsibilities—morning calendar briefs, ongoing research, reminders, and project monitoring—then manage them from the Dot profile.

Your Dot can use ChatGPT memory and its own saved context to learn your preferences, workflows, goals, standards, and previous feedback.

Where supported, you can continue the relationship across ChatGPT desktop, web, mobile, Slack, Teams, and voice—without starting over.

ChatGPT helps with a conversation. A Dot can own a responsibility.

Continuity is the throughline: a Dot can own a defined part of your workflow rather than repeatedly tackling isolated requests.

You control what a Dot can access, how independently it may operate, what needs approval, and when work should pause. Dots can make mistakes; consequential output and action deserve review.

• Custom Rules. Define which actions can happen automatically, only when pre-approved, after asking, or only by you.
• Auto-review. Potentially consequential actions are checked against instructions, permissions, Custom Rules, and safety requirements.
• Watch and redirect. Inspect In Progress, Scheduled, and Completed work; open the cloud computer and change course when necessary.
• Pause or reset. Pause background work any time. A reset deletes the Dot, its conversations, memories, and scheduled tasks.
• Personal computer stays separate. Local access starts off and requires explicit desktop-app connection and permission; it can be revoked later.
• Privacy and access. You decide which applications a Dot can use. Disconnecting an app does not necessarily delete information previously obtained from it.

Give it a job, not a vague prompt.

Define the outcome, available information, actions it may take independently, approval boundaries, what to monitor, cadence, what good looks like, and when to involve you.