Edit for your voice
Writing samples, a local style profile, a checked diff, and the optional AI-use statement.
In short
The voice editor learns how you write from samples you give it and proposes changes to a passage so it reads like you. Every change is checked in code before you see it — numbers, citations, math, commands and meaning words are protected — and you apply it hunk by hunk.
1Add writing samples
Settings → Writing voice: paste text or upload .txt, .md, .tex, .docx or .pdf. A style profile is computed locally, and you can edit it.
2Select a passage and press Edit for my voice
Works in Code and Visual mode, on a selection or the section at the cursor, on papers, proposals and the email.
3Apply or reject each hunk
A hunk that would change a number, a \cite key, math, a command, a comment, a reference, quoted text, a URL or a meaning word is dropped and reported. A one-for-one word swap with no rule behind it is never made.
4Optionally, generate an AI-use statement
Generate AI-use statement builds a short factual statement from the document's own record. It is inserted only where you put it, or copied. It is never automatic and never required.
For technical users
The rules are adapted from the owner's humanizer skill into a versioned voice-rule list; nothing targets AI-detector scores. Samples live in the app's user data, never in a project, and are sent only with a voice edit to the model you run it on. A kept hunk that removes an article and leaves a bare noun carries a "Check: missing word?" badge rather than being dropped. Traced claims keep their traces (re-anchored).