1
Download a model once
Pick a Whisper model from 78 MB. It's a plain file fetch with no account attached, and it lives on your phone from then on — that's the whole trick.
Private Transcribe
On-device AI · free on iOS & Android
Speech to text that never leaves your phone.
Whisper AI runs on the phone in your hand. Record a thought, a meeting, a lecture, and get accurate text in seconds, in 99+ languages, with no internet at all. No account. No cloud. No uploads, ever.
works in airplane mode
no account
99+ languages
Pro is a one-time purchase
Watch it happen
Tap. Speak. It's text.
A live waveform shows you're being heard; stop whenever, and the words appear within seconds. That entire round trip happens inside the phone, which is why it works just as well with the radio off.
Done · stays on this phone00:07
This is exactly what I wanted to say, captured word for word — and it never left my phone.
How it works
Three taps to text. Zero round trips.
One download, then everything runs in airplane mode. Here are the three screens you'll actually use.
2
Tap and talk
A live waveform shows you're being heard. Stop whenever you like; long recordings are welcome, and the screen stays awake while the model works.
3
Use your words
Edit, search, copy, share. Every transcript lands in a history grouped by date, with word counts and playback; Pro adds .txt and .srt export with timestamps.
Private by architecture, not by promise
There's no server to trust.
There's no server.
Cloud transcription services promise to protect the recordings you upload. We remove the upload. The speech recognition engine physically runs on your phone, so there is nothing to breach, leak or subpoena.
Your voice
The microphone is used only while you're recording, and the audio is written to the app's local storage.
Runs on your phone
Whisper AI, on-device
whisper.cpp runs the model on your phone's own processor. Airplane mode on, radio off — it still works.
Your text
Editable, searchable, yours. Copy it, share it, export it — or delete it. Nobody else ever had a copy.
No transcription server exists. Not ours, not anyone's.
Never leaves the phone
- Recordings — stored only in the app's local storage
- Transcripts and their file names
- The models you downloaded
The only things that touch the network
- A one-time model download from Hugging Face's public servers
- Ads in the free version (Google AdMob; non-personalised without consent) — gone entirely with Pro
- Purchase validation when you buy Pro, handled by Apple or Google
Your controls
- No account, no sign-up, nothing to leak
- Settings → Clear All Data wipes recordings, transcripts and models
- Deleting the app deletes everything — there's no server copy
Pick your accuracy
You choose the engine. Cloud apps don't let you.
Four tiers, each in an English-only version (fastest) and a multilingual one. Trade speed for precision per task, and switch any time.
Quick
78 MB
Transcribes in seconds. Great for short notes and quick captures.
Notes · captures
Recommended
Balanced
148 MB
The everyday sweet spot of speed and accuracy. Start here.
Everyday
Accurate
488 MB
Noticeably better. Shrugs off background noise and accents.
Noisy rooms · accents
Professional
1.53 GB
Highest accuracy available. For meetings, interviews and recordings that matter.
Meetings · interviews
99+languages
EnglishSpanishFrenchGermanPortugueseChineseJapaneseKoreanHindiArabic+ 90 more
Multilingual models understand all of them offline. With Pro, any of them can be translated into English text as you speak — also on-device.
Built for real work
Everything you said, searchable.
Meeting notes, interviews, lectures, voice memos, drafts, journals. Every transcript lands in a history grouped by date, so the thing you said three weeks ago is one search away.
- Search, edit and re-read any transcript
- Word counts and per-recording playback
- Copy or share in one tap — or auto-copy every result to the clipboard
- Screen stays awake through long transcriptions
- A dark interface built for long sessions
Meeting notesInterviewsLecturesVoice memosDrafts by voiceJournaling
The screens
Four tabs. Nothing to sign in to.
The whole app, in order.
First open
The promise, up front: your voice stays yours. No account to create, nothing to agree to but the mic.
Record
One button. The live waveform is driven by your real mic level; the timer counts wall-clock time.
Result
Your words, a word count, a green check — and Copy and Share right there.
Models
Eight Whisper models from 78 MB to 1.53 GB. Download, select, delete — and a storage check before the big ones.
History
Grouped by date, searchable, editable. Each entry remembers its model and duration.
Settings
Your stats, your defaults, and a storage panel where everything can be wiped in one tap.
Pro
Pay once. Never subscribe.
The free app already does the private part completely. Pro is a single purchase for the people who use it every day — and it's the only paid thing in the app.
$7.99one-time · no subscription
- No ads, ever — every banner and interstitial removed permanently
- Translate speech from 99+ languages into English text, on-device
- Export transcripts as .txt files or .srt subtitles with timestamps
- Restores on your other devices through Apple or Google
The free version shows a banner and, now and then, an interstitial after a transcription finishes — never while you record, never before your third transcription, and never twice within three minutes. Price shown in USD; your store shows the local equivalent.
Questions
The short answers.
Is my audio really never uploaded?
Never. Transcription runs on your phone's own processor via whisper.cpp. There is no account, no sign-up and no transcription server — the app works with networking fully disabled. Try it in airplane mode.
Does it work offline?
Yes. You need a connection once, to download a speech model (from 78 MB). After that, recording and transcription are fully offline — on a plane, in the field, in a basement lab.
What does it cost?
The app is free with ads. A one-time Pro purchase — not a subscription — removes all ads forever and adds translation into English plus .txt and .srt export.
Which languages are supported?
English-only models are the fastest. Multilingual models transcribe 99+ languages — Spanish, French, German, Portuguese, Chinese, Japanese, Korean, Hindi, Arabic and more — and Pro can translate any of them into English text.
Which model should I pick?
Start with Balanced (148 MB). Move up to Accurate or Professional for noisy rooms, accents or important recordings; drop to Quick for instant notes. Every tier comes in an English-only and a multilingual version, and the app checks free storage before a big download.
Do I need an account?
No. No sign-up, no login, nothing to leak. Everything lives in the app's local storage on your phone.
How do I delete my data?
Settings → Clear All Data removes every recording, transcript and downloaded model. Deleting the app does the same. There is no server copy to chase — there never was one.
If your voice is worth writing down,
it's worth keeping private.
Free. No account. Works in airplane mode.