Speak, and it becomes writing. All inside this Mac.

Deletion, not generation.

The core of cleanup is not a generative model — it’s our deletion tagger, which only marks words for removal. The output is always a subset of the input, so hallucination, summarising and meaning drift are impossible by construction.

In the modes that use Apple Intelligence, a verification guard discards its output the moment it detects substituted or invented words, falling back to the local result.

Built for spoken Japanese.

Fillers, false starts, stutters, recogniser quirks — a 40 MB model built solely for spoken Japanese removes them in about 4 ms. You won’t feel it at all.

Listening the instant you press.

Recording starts the moment you press, so the first word is never clipped. You see a live preview as you speak — and the instant you release, finished text is already at your cursor.

You decide how much to polish.

Five levels, from “don’t touch a word” to “document-quality rewrite.” The trade-off between fidelity and readability stays in your hands.

Only Rewrite changes wording (outside the subset guarantee). Two-pass uses Apple Intelligence for punctuation, line breaks and lists only — its output is discarded if the guard detects substitutions.

Small things that add up.

Requirements and FAQ