New in version 1.7: On-device processing (beta) On devices with Apple Intelligence (iPhone 15 Pro or later, iOS 26 or newer), the app can now transcribe and format recordings right on the device, with no OpenAI key. Once the language data has been downloaded, it also works offline. Turn it on in Settings by choosing "On-device (Apple Intelligence, Beta)". OpenAI stays the default, and the app never switches between the two on its own. The on-device model is smaller than GPT. If a dictation contains a question or a request, it may answer it instead of only rewriting it. When the app notices a made-up result, it shows your raw transcript instead and lets you try again. Very long recordings can't be formatted on the device; your raw transcript is kept and can be shared. The share extension can now format text on the device too. New models The app now supports OpenAI's new GPT-6 models (Luna, Sol and Astra). If you never picked a model, the app now uses GPT-6 Luna. OpenAI retires GPT-4o and GPT-4o mini in January 2027; a model that is no longer offered is replaced by the default, and Settings tells you. Reliability Recordings are kept until they are safely saved to the history. A failed recording can be retried from the last completed step, so a transcript that already exists isn't created again. On the watch, the app now checks on its own whether a recording has finished processing and then removes the "waiting" message. The diagnostics report also checks whether recordings arrived in the history.

New in version 1.6.1: Reliable watch recordings A recording made on your watch now stays stored there until your iPhone has explicitly confirmed that it arrived. If the connection drops, the iPhone is locked, or either app is terminated, the transfer resumes on its own later instead of losing the recording. Recordings still waiting to be processed are now visible on the main screen. A recording that failed is also no longer pushed aside by the next one. It appears in the same list as the ones from your watch, where you can retry or discard it. Two status messages on the watch used to linger after they no longer applied. That is fixed. Under Settings › Diagnostics you can generate and share a report about the transfers. It contains no recordings, no text and no API key.

New in version 1.6: Apple Watch Recording can now start automatically as soon as you open the app on your watch. Turn it on from your iPhone under Settings › Apple Watch › "Start recording when opened". There is also a complication for your watch face: tapping it starts a recording right away, regardless of that setting. And a recording in progress can now be discarded instead of having to be sent. Very short recordings are discarded automatically. The watch app is now fully translated and shows the correct app name everywhere. Recordings from your watch and via "Share" now get automatic titles and tags as well. Fixed The share extension was missing from the share sheet on many iPhones, and the watch app could only be installed on the very latest watchOS version. Both were caused by system requirements that had been set too high, and both are corrected. Settings have been reorganised: instead of one long list, an overview now leads into the individual sections.

New in version 1.5: Recordings from your Apple Watch are no longer lost when processing fails. Until now the recording was deleted in that case. It is kept now, and VoiceLog automatically makes a second attempt. If that attempt fails too, the recording appears on the main screen under "Not processed". You can retry or discard it there for a week. The failure notification now offers "Try again" directly as well. The settings for automatic titles and tags are now fully translated.

New in version 1.4: SpeakDraft now automatically generates a fitting title and tags for every recording. Based on your transcript's content. You can filter and search your history by tags to find old recordings faster. Enable it in Settings under "Auto-generate title & tags."

New transcription model Transcription now uses GPT Transcribe by default. It is more accurate and cheaper than the previous default model. Better language fallback in unsupported countries If the device's system language isn't supported, the app now shows English instead of German as the fallback language.

More ways to get polished text Sometimes a voice recording isn't an option. This update lets you type or paste text directly and format it from any app on your phone. What's new: Text input: Type or paste text and run it through any of your modes, no recording needed. Share Extension: Select text in Safari, Mail, or any other app, share it with VoiceLog, and get the formatted result back instantly. Shortcuts: The new "Format Text" action turns VoiceLog into a building block for your automations. Personal context: Describe yourself once. VoiceLog adds this to every prompt automatically, making all modes feel more tailored to you. Model per mode: Custom modes can now use their own GPT model, independent of your global setting. Prompt tester: Test a custom mode with a sample text right inside the editor before saving. Cost overview: See at a glance what the current month has cost you so far.

New in Version 1.1: • Apple Watch: Dictate from your wrist, result goes straight to the clipboard • Lock Screen widget for instant recording • Action Button support (iPhone 15 Pro+) • Share button on results (AirDrop, Mail, Messages …) • Output language: Spanish, Italian and French added • App UI now available in Spanish, French and Italian