Effective 24 September 2026
OpenAI creates fictional narration and dialogue from your messages, the selected character and scene, relationship context, and active memory summaries you can see and edit. Permission to share this information is requested before the first AI reply and can be withdrawn in Settings. Declining leaves AI replies unavailable; account controls and legal information remain accessible.
Automated input and output filters check the supported English/Latin-script content path. Rejected input creates no provider call or stored turn. Generated replies are checked as complete turns before display and storage. These filters have limitations and may block harmless text or miss unsafe content; they are not a guarantee that every conversation is safe. Minor self-disclosure ends access. In a crisis, use real-world support rather than this fictional app.
We do not supply other users' conversations as your story context or use your conversations to train our own models. Replies are generated through the OpenAI API with store: false, so requests are not stored for later retrieval. OpenAI does not use API data to train its models by default and may retain API data for up to 30 days for abuse and misuse monitoring. If generation is interrupted, the app may show a short, prewritten fallback; it is not a reply from a person.