Whisper, a third engine that runs every language from one model you pick, on any iPhone. - Whisper joins SpeechAnalyzer and Dictation. Pick one of five models, from Tiny to Large Turbo, and one download covers ninety-nine languages, including ones Apple's engines don't offer. Your words appear after you stop. - Faster with the Neural Engine: an optional second download lets Whisper run on the Neural Engine. - Downloading a model is the only thing the app ever does over the network. It comes from one fixed host, picks up where it left off if interrupted, and sends nothing. Everything else still works in airplane mode. - Long recordings are transcribed in pieces, so an hour costs the phone no more memory than a minute, and the entry shows how far along the run is. - While Whisper works, a new entry holds its place with a shape the size of the words to come. - Switching language mid-recording now cuts at the pause around your switch, so the words on each side land whole and in their own language. - A new transcription screen: your language on top, a switch between the three engines, and a line on each saying when your words appear. - The first run asks when you want your words to appear, as you speak or after you stop, and picks the engine for you. - The Cache screen shows what the app keeps in one bar, models included. - VoiceOver names every model control.

Internal cleanup. No changes to how the app works.

Faster everywhere, a clearer backup screen, dictation that keeps what you said before a pause, a new first run, and polish across the app. - Speed, everywhere the journal is read: the app no longer re-decrypts your whole journal after every rename, edit, delete, or landed re-transcription, and it decrypts on a worker thread before the first read, so a long journal stays as quick as a short one and the list is there when the splash lifts. Scrolling home, the live transcript, and reflections all stop redoing work they had already done. - The backup screen redesigned: it opens by saying what a backup holds and roughly what it weighs, each format is its own action, Export as Markdown, Obsidian Vault, or Website, with one switch for whether recordings ride along, and packing runs off the interface thread with a live percent and a cancel. Passphrase fields can be revealed while you type them. - Restore is honest about what it does: an entry that already exists takes the backup's version, you see what the backup holds before you commit, and the summary afterwards splits what was added, what was replaced, and what was already there. Failures name their cause instead of guessing, and the Website export speaks your language rather than shipping English to every locale. - Onboarding reworked as four short pages built from the app's own parts instead of a list of icons: a take running with the recorder's own wave and live text, your week read back (on iPhones with Apple Intelligence), yours in any shape (the export formats and the sealed backup), then set up, where reflections, the microphone, speech recognition, and reminders sit as live rows and Get started asks for them, so saying yes to reminders there turns them on. You can swipe back and tap the dots; forward stays on the button. A permission you denied offers Settings where you hit the wall and the first entry you open points once at its menu. - Dictation keeps what you said before a pause. Stopping for a couple of seconds used to make the recognizer start over, and everything before the pause was dropped from the live text and from the saved entry; every sentence of a take is now kept, in order, once each. Re-transcribing a long entry no longer stalls out, an entry whose recording is gone says so instead of failing generically, and live text in Japanese and Chinese wraps across lines instead of running off the edge. - Polish across the app: the glass bar buttons answer a tap in the hand like every other control, and so do a scrub and a tap on an entry's wave. Editing a transcript, a title, or a passphrase now works the way iOS text does, with a tap to place the caret, the magnifier, handles, and the edit menu (so a password manager can paste); a transcript and a title capitalize their sentences, and dragging the page puts the keyboard away. Changing the app language moves the reflection reminders to the new week and words at once. The app stays portrait. VoiceOver names every bar button, floating disc, switch, and segment, and hears notices instead of watching them fade; faded live text and inert calendar days read at 3:1 in every theme, and follow Increase Contrast.

Say more into an entry you already saved, re-hear the whole journal with a better engine, and dress the app in a look of your own. - Continue an entry: record more onto a saved entry from its own screen. The audio merges into the kept recording and the transcript grows to match, edits kept, and the new words shimmer into place where they land. A take in another language gets the language marker. An entry with no recording can be continued too; the take becomes its recording. A take that cannot be merged is saved as its own entry, never lost. - Re-transcribe all: a bulk runner over the whole journal, seated under the engine picker on the transcription screen, that lets the engine you switched to re-hear the entire history. Live animated progress with cancel, and resume is free: a run skips whatever the current engine already transcribed, so stopping loses nothing. It waits for a live recording and pauses while the device runs hot. - Every feature is free, including the formatted exports the club used to gate. What the OpenTranscribe Club unlocks from here on is looks, and nothing else. - Club themes: every family beyond Default is a club look. Gruvbox and Sepia stay, and five join them: Midnight, Dracula, Nord, Catppuccin, and Tokyo Night. Solarized is retired. A pick is kept whether or not the club covers it, so it snaps on the moment a membership lands. - Club app icons: three alternate home screen icons, Signal, Lines, and Dots, picked on the appearance screen above the themes. A club look, like the families. - The appearance screen holds both pickers now: the app icons over one grid of every theme, with light, dark, and system moved into the bar's own menu. Tapping a club look you have not joined for opens the club, and joining wears the look you tapped. - The club opens as a sheet over whatever you were doing, rather than taking you to a screen of its own.

Backups, editing, the club, engine choice, and faster ways to start a recording. - Formatted exports, the first OpenTranscribe Club feature: the whole journal handed to the share sheet as Markdown, as Obsidian notes, or as a Website that opens with a player in any browser. - Native archive backup: one file that restores the whole journal, audio and reflections included, sealed with a passphrase by default; without it the file is unreadable. Free for everyone, never behind the paywall, so anyone can always back up and recover. - The OpenTranscribe Club, an optional one-time purchase that supports the app: club features start with the formatted exports, and whatever joins the club later is included. Direct StoreKit 2 with no purchase SDK, no account, and no server: the entitlement is verified on-device from Apple's own record, membership works in airplane mode, and no journal content is in the purchase conversation. - Manual transcript editing, in place on the entry screen, with a revision history that keeps every prior version and lets one be restored. - A lock screen control and a widget row that start a recording without opening the app, plus Siri and Shortcuts support through App Intents. - The launch splash is now drawn natively and plays over the boot, replacing the earlier Flutter splash screen. - Notifications: a master switch for reflection reminders, a toggle per period (day, week, month) for which ones nudge, and one shared time. - Engine choice: the transcription screen lists every engine the app ships, SpeechAnalyzer (the iOS 26 engine) and Dictation (the classic recognizer the system's dictation uses), each described in one line and switchable with a tap. An engine the device cannot run stays visible, dimmed, with the reason; a new engine lands as one more row. - The transcription screen redesigned around the language you speak: the default language is a card with an honest status line (ready and naming the engine that answers, download progress, or what stands in the way), the other ready languages are chips a tap makes the default, and the whole library moved into a sheet where languages are added, removed, and switched in one place. - iPhones whose hardware cannot run SpeechAnalyzer (the iPhone 11 family and earlier chips) no longer show an empty language list: they start on Dictation and transcribe normally. - Under Dictation, a language whose system dictation model is missing says so and points at the iOS keyboard settings, instead of claiming it is ready. - Reliability around engine switches: a switch landing mid-download, mid-removal, or mid-load can no longer leave one engine's status, failure badge, or download progress on the other engine's language rows; queued model downloads survive a cancelled predecessor; and a blocked install whose slot-holders have since left offers a retry instead of a dead end.