Tap once and talk. OpenTranscribe records, transcribes, and reflects entirely on your iPhone. No account, no sync, no telemetry, and it works the same in airplane mode.
Met Lia for coffee and we ended up talking about the move for two hours. I keep saying I want a smaller life and then filling every evening. Walked home the long way. The city was quiet for once. Then I sat on the steps for a while and did nothing, which felt like the point. Work was fine. Nobody asked for anything I could not give. Tomorrow I want to call Mum before it gets late.
A handful of screens, and none of them needs a connection.
Tap once and talk. A live transcript appears while you speak, and when you stop, the full recording is transcribed on the device before it lands on home.
Apple speech models are downloaded once per language, or one Whisper model serves them all. Either way they run entirely on the handset, and airplane mode changes nothing.
An entry holds the transcript beside its recording, organized by day across the week. Audio is kept by default, so when a better engine ships, old entries can be transcribed again and read better than the day they were spoken.
Keeping audio is a preference. Turn it off and each recording is deleted the moment its transcript lands.
Apple Intelligence reads the entries and writes a short reflection for every day, week, and month, entirely on the device. Voice, length, and how specific it gets are yours to set.
A quiet week is a valid result. When there is little to say, it says little, or nothing at all.
Home is the week: entries grouped by day under the week strip, reflections folded in above the days they describe. Nothing to file, nothing to tag; the calendar is the structure.
The strip moves between weeks. Tapping the title always brings it back to today.
Each language runs its own Apple speech model, downloaded once and shared with the system, or one Whisper model of your choosing serves them all. Pick the languages you speak and recognition follows, with no server behind it.
Engines are swappable behind one contract, and every one has to declare it runs on the device before the app will load it.
Pick when your words should appear and the app picks the engine. Every engine has to declare that it runs on the device before the app will load it.
OpenAI, through whisper.cpp
An open model on whisper.cpp, one download for every language. Writes your words after you stop.
32.2 MB · Basic
The smallest and fastest. Rough on names and accents, fine for a quick note.
59.7 MB · Good
Quick, with clearer words than the smallest. Good for short notes in a quiet room.
190.1 MB · Better
The balance most iPhones want. Accurate for everyday speech in every language.
539.2 MB · Best
Slower and careful. Strong on accents, quiet takes, and long entries.
574.0 MB · Top
The best on offer. Needs a recent iPhone and a little patience per entry.
Apple
Apple's newest engine, a downloaded model per language. Shows your words as you speak.
English (US)
Ready · SpeechAnalyzer
Apple
The recognizer behind iOS keyboard dictation. Shows your words as you speak.
English (US)
Ready · Dictation
OpenTranscribe
Club
Everything that makes OpenTranscribe useful is free for everyone, and stays that way. The club is how the app is supported: an optional one-time purchase, with a few looks as thanks.
It is made directly through the App Store, with no purchase SDK, no account, and no server behind it. Membership is verified on the device from Apple's own record, so it works in airplane mode, like everything else.
$25
Once. No subscription.
In US dollars; the price follows your App Store region.
What you get
Signal, Lines, Dots, and every icon beyond Default.
Default
Signal
Lines
Dots
Every family beyond Default, each in light and dark.
Gruvbox
Midnight
Dracula
Sepia
Nord
Catppuccin
Tokyo Night
A claim like that cannot be backed by a privacy policy, only by source you can read. The code is public and MIT licensed, and every card names the file that holds it, linked to the source.
One allowlisted file, one pinned host, nothing sent. A test fails if any other file reaches the network.
What the app connects to
Every engine answers onDeviceOnly. The app refuses one that answers false.
Only paths, durations, levels, and text cross a channel. Audio bytes never do.
A random per-device key, generated on first launch and held in the Keychain.
Direct StoreKit 2, verified on the device. No purchase SDK, no server.
No Swift code reaches a method by name at runtime, and a test bans it.