Nothing you write here is used to train any model. Not your script, not your characters, not your notes. Not now, and not later.
Your script is stored so you can open it again, and it is sent to a model only in the moment you ask a question — when you press Analyze, ask the room something, or ask it to propose a value on a character. It is not kept on a server between those requests, and it is not collected anywhere for training.
We use the Anthropic API, which does not train on what is sent through it. That is their commitment as well as ours.
The craft knowledge in northsetpro comes from published screenwriting books and from a corpus of 106 produced, publicly available screenplays that were read for structure — where beats land, how arcs are shaped, how motifs recur, how scenes carry subtext. What the tool holds is derived analysis: positions, patterns and counts. It holds no screenplay text, and there is an automated check that fails the build if any ever appears.
Facts about how a film is built are not the film. That distinction is the whole basis of the knowledge base, and it is the same one that lets a screenwriting book discuss Chinatown for forty pages.
A scene you cut is kept in your own graveyard so you can get it back — visible only to you and the people you have shared the script with. Deleting a script removes it and everything under it.