Let Sudowrite Learn from What I Actually Insert

Currently, once I insert and refine a passage, Sudowrite more or less forgets it. It’s now there within the larger context that Sudowrite sees, but essentially every new generation starts fresh, using only the prompt and style instructions I’ve explicitly given it elsewhere. However, the tool now has something that seems far more valuable: the version I decided was best, with my edits, from my specific prompt.

Seems like a missing opportunity for dialing in the AI’s writing voice.

I suggest an optional toggle that allows Sudowrite to learn from the text I ultimately insert.

When enabled:

• Sudowrite would store the original prompt + context and the final inserted text (including my edits).

• Future generations in the same project would reference these accepted examples to better match my voice, tone, and intent.

• This would be project-scoped and opt-in, with clear controls to reset or remove learned examples.

In addition to getting better and better outputs, the non-obvious benefit is that Sudowrite would be learning what I mean when I prompt — e.g., what *“ominous,” “uncanny,” “restrained,” or “subtext-heavy” actually look like in practice (for me).

To take it a step further — using this feature could essentially create a new curated “writer” to use for specialized generation (i.e., character or purpose-specific). The user could easily toggle between these “writers” for different purposes, the most obvious use case being in the training of different POV voices. (And in a perfect world, the AI would figure out which “writer” to use on its own, based on context.)

But that’s all a bit specific and nuanced for the tl;dr: let the AI learn from what I choose to ultimately use from a given prompt.