Short answer
Give it one app for one job. Use action rules to decide what it can do on its own, and what needs your say-so. Want a clean slate? Reset the Dot; disconnecting one app only stops future access.
How action rules work
Custom Rules can set boundaries for supported actions, such as telling a Dot not to send email. They cannot override built-in safety requirements or Auto-review. “Pre-approved” applies only to the action you clearly authorized; it is not blanket permission for future actions. Read the safety guide before setting rules. OpenAI privacy and safety FAQ
Optional separate Slack account
OpenAI says you can set up an optional separate Slack account for your Dot, giving it its own identity and access in Slack. This does not change the plugins connected through ChatGPT’s Plugins tab or their shared permissions: the Dot can still use those plugins with its own Slack account. OpenAI privacy and safety FAQ
What happens between chats
A Dot can check connected apps and save memories between chats. Handy for recurring work; a good reason to keep an eye on Activity. Need quiet? Pause is its clock-out button.
Connecting your computer
Computer access starts off. To switch it on, open the desktop app on that computer and choose Allow access. The Dot can then reach files through its messaging channels. With access enabled, it can create Work or Codex tasks, use local skills, and use your local browser when its cloud browser is blocked. Revoke permission in the app when you’re done. Its cloud computer is separate from yours. OpenAI Help Center
Codex cloud tasks
A Dot can create cloud tasks in Codex cloud environments. Set up the environment in Codex before asking your Dot to use it. This is separate from optional access to your local computer. OpenAI Help Center
Approvals and the pause control
Choose which supported actions need approval. Custom Rules cannot disable core safety requirements or Auto-review. Actions such as changing a password or transferring money require you to take over; deleting data or installing software may require approval each time. Any advance approval stays limited to the specific action you described. You can pause from the profile menu; pausing does not erase data. OpenAI privacy and safety FAQ
What happens when you disconnect an app
Disconnect stops new access through that connection; it does not delete information already built into the Dot’s context. OpenAI’s privacy and safety FAQ explains how Dot context, ChatGPT Memory, and separately stored files differ. OpenAI privacy and safety FAQ
Reset: what gets deleted
Deleting your Dot removes its own context. Files, Codex threads, and ChatGPT conversations it created are stored separately and are not deleted with the Dot. Memories from other ChatGPT conversations remain in ChatGPT Memory settings. Read the deletion notice before resetting. OpenAI privacy and safety FAQ
Memory and model training
A Dot can receive ChatGPT memories and recent conversation context, and Dot conversations can contribute to ChatGPT Memory. Turning off Memory stops future sharing, but does not remove information already received. Business, Enterprise, and Edu data is not used to train models by default; personal-plan training depends on “Improve the model for everyone.” See the full privacy and safety guide and OpenAI’s data controls. OpenAI privacy and safety FAQ
Before you connect an app
Connect an app to solve a specific chore, not just to see what happens. Check what it can expose and which actions need your approval. Start with one; expand only when there is a reason.
When the pause status and activity disagree
Added October 5, 2026. OpenAI documents Pause and Resume in the Getting started guide. Separately, a user report on GitHub describes a precautionary pause and failed recovery controls while new task progress remained visible. That report is not an OpenAI confirmation or an independent reproduction. Visible progress alone does not establish which background actions or external changes occurred; the affected versions, cause, and scope remain unconfirmed.
What to check before resuming
Use the supported activity and review controls to inspect the affected task. Check the destination app for actual changes, and keep a record of the pause message, times, app version, and recovery errors. A restart or Reboot request is not proof that execution stopped. If the controls fail or status stays inconsistent, contact support with those details and avoid assigning more consequential work until the state is clear. Do not bypass Auto-review. Reset deletes the Dot’s conversations, memories, and scheduled tasks, so it is not a harmless troubleshooting step.
FAQ
Can OpenAI Dots access my computer?
Only if you opt in and confirm Allow access in the desktop app. Computer access starts turned off and can be revoked.
Does disconnecting Gmail or Slack erase what Dot already read?
No. Disconnecting stops new access through that connection; it does not delete context already built. Deleting the Dot removes its own context, but not separate files, Codex threads, or ChatGPT conversations it created.
Can Custom Rules override safety checks?
No. Rules can set boundaries for supported actions, but cannot turn off core safety requirements, Auto-review, or proactive research restrictions.
Are business Dots used to train OpenAI models?
OpenAI says it does not train on ChatGPT Business, Enterprise, or Edu workspace data by default. Personal-plan controls depend on the “Improve the model for everyone” setting.
Can I pause a Dot?
Yes. Open its profile menu and select Pause. Resume from the paused state when ready.