LocalWispr turns your voice into polished text in every app — email, chat, docs, code. The AI runs on your Mac's Neural Engine. No cloud. No account. No audio upload, ever.
Free to try · Apple Silicon · macOS 14+ · Notarized by Apple
Hold Fn, speak, release — clean, punctuated text appears at your cursor, in any app.
Mail, Slack, iMessage, Notion, your editor, your terminal. If you can type there, you can speak there.
"Meet at six — actually, five" becomes "Meet at five." Spoken lists become bullet points. Filler words vanish. All on Apple Intelligence, all local — and a safety check guarantees it never changes what you meant.
A live waveform and rolling transcript show exactly what's being heard — and a crossed-out reveal shows every edit the AI made.
Double-tap the hotkey to lock recording — dictate a whole email leaning back in your chair, tap once to finish.
A personal dictionary teaches it your names and jargon ("k eight s" → "k8s"). Snippets expand "my work address" into the whole thing.
Dictate in English, Spanish, French, German, Portuguese, Italian, Dutch, Polish, Ukrainian and more — automatically detected.
Transcription runs ~190× realtime on Apple Silicon. Your words land before cloud apps would finish uploading the audio.
One model download at setup (~1 GB), then the internet is optional. Forever. Try it with Wi-Fi off — nothing changes.
Cloud dictation apps stream your voice — and often the text on your screen — to their servers, then share transcripts with third-party AI providers. LocalWispr was built so that's impossible: there is no server.
| LocalWispr | Cloud dictation apps | |
|---|---|---|
| Your voice uploaded | Never — processed on your Mac | Every dictation, streamed live |
| Reads your screen | Never | Often — window text sent as "context" |
| Works offline | Yes, completely | No — outages stop your dictation |
| Account required | None | Yes |
| Analytics & telemetry | Zero — no tracking code exists | Extensive |
| Your transcripts live | In plain files on your Mac | On their infrastructure |
| Price | One payment, yours forever | $12–15 every month |
Download free and evaluate as long as you like — the app is fully functional. If it earns a place in your day, buy a license.
Checkout page coming soon — for now, email and a human sends your key.
After a one-time speech-model download (~1 GB from Hugging Face at setup), yes — every transcription, every AI edit, every feature runs on your Mac's Neural Engine. Block the app's network access or turn on Airplane Mode: nothing changes. That download is the only network request the app ever makes, besides checking for updates when you ask it to.
Any Apple Silicon Mac (M1 or newer) on macOS 14 Sonoma or later. The optional AI polish feature uses Apple Intelligence, which requires macOS 26.
LocalWispr uses NVIDIA's Parakeet speech model, which outperforms Whisper large-v3 on public English benchmarks — with punctuation and capitalization built in. A personal dictionary handles your names and jargon.
Input Monitoring lets the app notice the hotkey being held in any app — it's a listen-only tap that architecturally cannot log or alter your typing. Accessibility lets it type the transcript at your cursor. The app's Settings shows a live diagnostic of both.
Audio is transcribed in memory and discarded. Transcripts are saved to a local history file only if you keep that setting on — you can search, export, or delete it anytime. Nothing is retained anywhere else, because there is nowhere else.
The evaluation is free and unlimited, so you'll know before you pay. If you buy and regret it anyway, email within 30 days and you'll get your money back.