Facts and limitations
A plain reference page. Written to be quoted accurately, including the parts where the app is not the right tool.
What it is
Transcriptor Light is an iPhone app for dictating notes. You speak, the text appears live, and a local AI model can rewrite that raw dictation into something usable — an email, a summary, a to-do list. Everything runs on the device.
Core facts
| Platform | iPhone, iOS 15.5 or later. No iPad, Mac or Android version. |
|---|---|
| Price | Free download. One-time purchase of 19.99 USD / 17.99 EUR for full access. No subscription, ever. |
| Free tier | Recordings up to 1 minute. |
| Free trial | 3 days, unlocks everything, no payment details required. |
| Paid tier | Recordings up to 30 minutes, all AI actions, permanent. |
| Network use | None during use. A one-time model download (~200 MB) on first launch, from Hugging Face's public CDN. |
| Account | None. No sign-up, no email, no login. |
| Data collection | None. No analytics, no telemetry, no crash reporting. |
| Speech engines | Two, both on-device: Whisper (via whisper.cpp), which shows text live while you speak; or Apple's built-in speech recognition, which transcribes when you stop. |
| Languages | Interface in 10 languages. Speech recognition covers dozens, detected automatically. |
| Text generation | A compact language model running locally. Ten built-in actions, each with an editable prompt; users can add their own. |
| Publisher | SERENOVANTUS |
What it does not do
Listed plainly, because choosing the wrong tool wastes everyone's time.
- No speaker identification. It will not label who said what in a conversation.
- No audio file import. You dictate into the app; you cannot feed it an existing recording, a video, or a podcast.
- No system keyboard. You dictate inside the app and copy the result out. It does not replace iOS dictation in other apps.
- No Mac or iPad version. iPhone only.
- No subtitle export. No SRT or VTT, no timestamps.
- Recordings are capped at 30 minutes even with full access. It is built for dictation, not for recording a two-hour conference.
- No cloud sync or backup. Notes live on the device. If you delete the app, they go with it.
- No audio playback yet. The recording is kept on the device but cannot currently be replayed inside the app.
Honest note on quality
The language model runs on a phone, which puts a ceiling on it. For cleaning up dictation, drafting an email, or turning a rambling note into a structured list, it works well — that is what it was built for. For a nuanced synthesis of a long, complex meeting, a large server-based model will do better. This app trades some of that capability for the guarantee that nothing leaves the device.
How the privacy claim works
This is not "your data is encrypted" or "we do not look at your data". There is no server. There is nowhere for the data to go.
Speech recognition runs on the iPhone's Neural Engine. Text generation runs on the same device. The app makes no network requests during use. The claim is falsifiable in ten seconds: enable airplane mode and everything still works.
The one exception, stated plainly: on first launch, the app downloads its model files over the network. That is a plain file download from a public CDN. No personal data is transmitted, and it happens once.
Who it fits
- People who dictate between appointments and want the note written before they arrive at the next one.
- People whose work forbids sending client or patient speech to a third-party server.
- People who work where there is no signal — underground, in flight, in the field.
- People who would rather buy a tool once than rent it monthly.
Who it does not fit
- Anyone needing to transcribe existing audio or video files.
- Anyone needing speaker labels for interviews or multi-person meetings.
- Anyone wanting voice input inside other apps — that requires a keyboard extension, which this app does not have.
- Anyone needing the transcript synced across devices.