OpenAI says disconnecting an app from a Dot stops new information sharing through that connection, but it does not erase context the Dot has already incorporated. A Dot is a persistent ChatGPT agent that can work between conversations using connected apps, so disconnecting, turning off ChatGPT Memory, and resetting the Dot have different effects. OpenAI’s setup guide explains the connection and reset controls.
What happens when you disconnect an app
Disconnecting an app stops the Dot from receiving new information through that connection. Information already incorporated into the Dot’s context remains while you keep the Dot. Turning off ChatGPT Memory also stops further sharing of memories, but does not remove information the Dot has already received. OpenAI’s privacy and safety FAQ describes these separate controls.
Resetting a Dot deletes its conversations, saved memories, and scheduled tasks. Files and conversations stored separately in ChatGPT, Codex, or other locations have their own retention rules, so resetting the Dot does not delete every separate copy.
| Action | What changes | What remains separate |
| Disconnect an app | Stops new sharing through that connection | Context the Dot already incorporated |
| Turn off ChatGPT Memory | Stops further memory sharing with the Dot | Information the Dot already received |
| Reset the Dot | Deletes the Dot’s conversations, saved memories, and scheduled tasks | Separately stored files and conversations, which follow their own retention rules |
Treat disconnecting as a stop to future sharing, not a delete button. Reset the Dot to remove its own context, and manage separately stored files and conversations where they are kept.
Model improvement and human review are different
On personal plans, the “Improve the model for everyone” setting controls whether eligible Dot conversations and work may be used to improve models. OpenAI says Business, Enterprise, and Edu workspace content is not used for model training by default. Background research threads and notes are not used directly, though information brought into an eligible conversation or task may be used depending on the user’s settings.
Turning off model improvement does not rule out limited human review. OpenAI says review may still occur in some circumstances, including for safety-related reasons. Model training and human review are separate practices, governed by different conditions. OpenAI’s FAQ describes both.
What proactive research can do
OpenAI says a Dot’s proactive research uses read-only tools: it cannot directly send messages, change connected-app content, or control a browser or computer. Any later action follows the usual rules and safety checks. OpenAI also says its safeguards reduce risk but do not prevent every mistake or prompt-injection attempt. Its safety explanation outlines those limits.
For supported secure sign-in, OpenAI says credentials entered into the sign-in form stay out of the model’s context. Passwords or other secrets placed in readable chats, documents, or plugins fall outside that protection.
For more on the difference between background research and follow-up actions, read NeoTeo’s earlier explanation of how Dots separate research from actions.
US access and texting
OpenAI says Pro access is for users 18 or older in eligible markets, with a gradual rollout. Texting is a separate limited beta for some US Pro users; it is not available in Business or Enterprise workspaces. The texting beta is not a general feature for every Dot user. OpenAI’s setup guide gives the plan and rollout conditions.
