VoiceDebrief for iPhone and iPad

A voice note in, a transcript and a short debrief out, on the device. The app is being built to do VoiceDebrief's one job with Apple's own models and nothing else: no OpenRouter, no API key, no inference server. What it produces is kept in an encrypted vault the user owns, opened with a passkey, and signed with the user's own key.

In development. There is nothing to install yet. The repository holds an app that builds and tests, the vault engine it will use, and the plan. It does not record, transcribe or summarise anything yet. What is built is the honest list, and it is short.

What it will do

StepHowStatus
Bring a voice noteRecord in the app, or share one from WhatsApp or Files (.opus, .ogg, .m4a)proposed
TranscriptApple's SpeechAnalyzer, on the device, with the time of every lineproposed
DebriefApple Intelligence, on the device: key points, actions, one sentenceproposed
Keep itAn sgit vault the user owns, encrypted before anything leaves the deviceproposed
Sign inOne passkey, the same account as pki.sgit.ai, and a recovery codeproposed
One identityOne key bundle (RSA-OAEP 4096 and ECDSA P-256) that signs every debriefproposed

What exists today

Status
The appAn iPhone and iPad app with one screen that says what is and is not builtbuilt
The vault engineSGitVault: reads and writes sgit vaults and derives a passkey's door, with tests against published vectorsbuilt
CIEvery push is compiled and tested on a Mac; the vault engine's output is read back by sgitbuilt
This siteBuilt from markdown, validated, and published from the dev branchbuilt

Where your audio will go

Nowhere, for transcription and the debrief. Both run on the device with Apple's models. That is the design, and it is why the app needs no key and no account with us.

Your vault is the one thing that leaves the device, and it leaves encrypted. Files are encrypted on the device before upload, with keys derived from a vault key the server never sees. The server can see the vault's identifier, the size of each encrypted object and when it was written. It cannot see file names, contents or keys. sgit's security model states this precisely.

None of this is built yet, so none of it has been tested on a real device. This page will say so when it has.

The record

Part of voicedebrief.ai, alongside MeetScribe for Google Meet.