One recording.
One summary.
Then nothing.

What Advanced Privacy is

A lock sits in the top left corner of the main screen. Tapping it leaves the ordinary app behind and opens a black room with a green line through it, dark whatever your appearance setting says, so there is never any doubt about which mode you are in. Inside, the app does one thing: it records once, transcribes it, erases the audio, writes the summary you asked for, and offers it to you. Then it forgets the lot.

It is not a stricter version of the privacy the app already has. Just Transcribe has always done its thinking on the device, with no account, no analytics and no server to send audio to. What this mode adds is the erasing, and the ability to check the claim yourself: put the phone in airplane mode after the models are downloaded, and watch a full session run with the radios off.

How a session runs

Six steps, and the order matters: everything that needs the internet happens before the recording does.

1. Open the lock

   The lock is disabled while the main screen is recording or busy, because both would be reaching for the same microphone. Otherwise it opens straight away, on a screen that looks nothing like the rest of the app.

2. Bring the models in

   The mode insists on Large v3 Turbo, the most accurate speech model in the app at about 950 MB, and on a summary model being present. If you already downloaded one for ordinary summaries, that counts: it takes the heaviest one your device can actually hold, rather than making you fetch another few gigabytes.

   This is the only part of a session that touches the network. The checklist puts it first for exactly that reason.

3. Cut the connection, if you want to

   Entirely optional, and entirely up to you. Swipe into Control Center, tap the airplane, and come back. The screen reports honestly whether a route out exists at all, so this is a claim you get to verify rather than believe. Everything after this point runs on the chip in your hand.

4. Decide the outcome before you record

   You choose the kind of summary first: a summary, the key points, action items, or a free instruction of your own, along the lines of "turn this into a short email". You can also have speech in any language come out as English text, still fully on the device. Once the recording stops you are not asked again, because being asked would mean the transcript sitting around waiting for an answer.

5. Record once, then let it run

   Recording starts the moment you tap, and you get one. A call or Siri in the middle is survivable; the parts are joined back together, and if they cannot be, the session says so instead of quietly handing you half a transcript. Stop when you are done and the chain runs straight through without asking anything else.

6. Take it with you, or lose it

   The summary appears with the full transcript next to it. Copy it to the clipboard, or share it out as formatted text with the date, or as plain text. Those export files are built inside the same temporary folder, so an unshared file cannot outlive the session either. Leave without exporting and there is nothing to come back to.

Where the recording actually goes

"Private" is a word every app uses, so here it is item by item, taken from what the app actually does.

• ✓Never the library. A session writes nothing to your recordings, your folders or your search index. Your ordinary transcripts are untouched, and nothing new appears there afterwards.
• ✓A temporary folder, protected. The recording goes into one folder created for that session, with iOS file protection applied so it stays encrypted while the device is locked.
• ✓The audio goes first. It is erased as soon as the transcript has been read out of it, before the summary is even started. From that point a session holds text and nothing else.
• ✓Everything in one place. Cleaned-up audio, interrupted fragments, export files: all of it is written inside that same folder, so one delete really does take all of it.
• ✓Wiped three ways. When you close the screen, when you confirm erasing, and at the next app launch if the app was killed before it could clean up.
• ×No cloud, no backup, no sync. Nothing is uploaded, nothing lands in iCloud, and there is no copy on another device. There is no server involved at any point.
• ×No settings left behind. Even the language you pick for a session stays in the session, instead of quietly becoming your new preference.

What it needs before you start

What it deliberately does not do

The honest list, because a mode built on a promise should be clear about where the promise ends.

• ×No importing. A session records on the spot. Existing wav, mp3 or m4a files belong on the ordinary side of the app.
• ×No editing, speakers, folders or search. One pass, no branches. Everything the app can otherwise do to a transcript needs a transcript that sticks around.
• ×Nothing to resume. No draft, no autosave, no history. That is the feature, but it does mean a crash or an empty battery costs you the session.
• ×No live captions. The transcript is made after you stop, not while you speak.
• ×It cannot make the room private. It keeps the recording off your phone. Who is listening in the room, and what the law where you are says about recording people, is still yours to sort out.

For anything you want to keep, the ordinary side of the app is the right place: it is just as offline, and it lets you edit, label speakers, search and export at your leisure. This mode is for the recordings that should not exist afterwards.

Who reaches for it

People who record conversations that must not sit on a phone afterwards:

• a journalist with a source who agreed to talk on the condition that nothing is kept
• a doctor, therapist or lawyer who needs the gist of a conversation and none of the audio
• HR or a works council taking notes on a case that is nobody else's business
• anyone under an NDA, in a due diligence room, or in a meeting about people who are not there
• anyone who simply wants to see, once, that a transcription app can work with the radios off

Questions