Questions, answered honestly

No. Your voice is captured, transcribed in memory on your Mac's Neural Engine, and discarded. It is never sent anywhere and never written to disk. There is no server in the transcription loop. The only thing Vocelle ever downloads is the speech model itself, once — see the privacy page for the complete inventory.

Hold right-⌘ anywhere on macOS and speak. When you release, Whisper large-v3-turbo transcribes your speech on the Neural Engine, an on-device model cleans up filler and punctuation, and the result types itself at your cursor as real keystrokes — in whatever app you're in.

It's a full-size key under your right thumb that almost nothing else uses, so holding it doesn't collide with existing shortcuts. Holding a modifier alone also doesn't type anything by itself, which is exactly what you want from a push-to-talk key. Press, speak, release — no chord to remember.

Yes. Vocelle synthesizes keystrokes rather than pasting from the clipboard, so text lands wherever your cursor is — Terminal, VS Code, vim over SSH, a browser text box, a native app. If a cursor blinks there, Vocelle types there.

Everything works. Transcription, cleanup, history, dictionary — all of it runs on your Mac. There's nothing to connect to. The only thing that requires internet is the one-time model download on first run.

Apple Silicon Macs running macOS 26 or later. Intel Macs aren't supported — Vocelle runs the speech model on the Neural Engine, which Intel Macs don't have.

It's in development. We're not going to name a date we might miss. What we can say: Pro and Lifetime purchases cover the iPhone app the day it ships, so there's nothing to buy twice.

About 1.6 GB, downloaded once on first run. After that, Vocelle makes no further network requests. First run also compiles the model for your Mac's Neural Engine, which takes a few minutes — see support for what to expect.

After transcription, an on-device model cleans up your words — filler out, punctuation fixed, self-corrections resolved. The guardrail is a check on that model: if its output doesn't keep at least half of the words you actually said, Vocelle discards it and falls back to deterministic cleanup instead. The practical effect: the model can tidy your words but can never answer you or replace what you said. The features page tells the full story.

It's designed not to, and the guardrail enforces it. Cleanup is limited to filler, punctuation, and self-corrections; anything that drifts beyond that gets thrown away. If you'd rather have the raw transcript with no model involved at all, your exact words are always preserved in history first — the write-ahead log saves them before any processing happens.

In local files on your Mac — nowhere else. In the history view you can search everything you've dictated, delete individual entries, or clear it all. Deleting removes the entry from your Mac; there's no server copy to chase, because none exists.

No. Audio exists only in memory during transcription and is discarded immediately after. It is never written to disk — not as a cache, not as a temp file, not "for quality purposes." Only the text is kept, in your local history, until you delete it.

None. No analytics, no telemetry, no crash reporting, no identifiers. We genuinely don't know how often you use Vocelle. The full accounting is in the privacy policy — it's short, because there's nothing to disclose.

Unlimited local dictation — the core of the product, with no word caps and no trial clock. It stays free. Pro adds history, the personal dictionary, hands-free mode, and the iPhone app when it ships. See pricing for the exact split.

A list of names, jargon, and terms you use that a general speech model wouldn't reliably get right. Vocelle uses it to spell your words your way. If you've built up a dictionary in another dictation app, you can import it in one click instead of retraining a new tool from scratch.

Dictation without holding a key. Toggle it on and Vocelle listens until you toggle it off — useful for longer passages or when your hands are busy. The live waveform overlay stays visible the whole time, so it's always obvious when Vocelle is listening.

You keep your words. Every dictation is written to a crash-safe write-ahead log before any processing happens. If Vocelle — or the whole machine — dies mid-dictation, the transcript is recovered from the log on next launch.

Microphone, to hear you. Accessibility, to synthesize keystrokes at your cursor and to detect the right-⌘ hold system-wide. Those are the only two permissions it asks for, and both are granted by you in System Settings. Support walks through granting them.

If you bought directly from us: email support within 14 days and we'll refund you, no questionnaire. If you bought through an app marketplace, refunds go through that marketplace's process. Full details in the terms.

vo-SELL — two syllables, stress on the second. Rhymes with "hotel."