← Back to the start
What is new
Every release in order, the newest first. Updates install from the app menu: Check for updates.
0.13.3November 2, 2026
The other side is recorded at the right speed
- The other side of a call could come out sped up and unintelligible, an hour squeezed into twenty minutes. The rate came from a format declared when the call started, and headphones change it midway when they switch to headset mode.
- The rate is now measured from the sound that actually arrives and follows the device while recording, so a change of mode is corrected within a second and a half.
- Calls already recorded are repaired before transcription: the microphone track is the true length of the conversation, and the other side is stretched back to it. That does not return the discarded samples, but the speech is intelligible again.
0.13.2November 2, 2026
A recording language is decided by the whole recording
- In Auto mode the language of a track was decided by its first audible piece. A short yeah at the start of a call could send the whole track into another language, and an hour of speech came back as stock subtitle phrases.
- The language is now decided by a vote over three samples from different parts of the recording, and not forced at all when there is almost no speech.
- The silence gate is higher, so a breath no longer reaches the model, and a reply that is one phrase repeated six times is dropped as a decoder loop rather than speech.
0.13.1October 30, 2026
The sound stays when you asked it to
- A recording transcribed later, from the waiting list, lost its sound even with Keep the audio of calls turned on. The decision lived only in the app memory, and a later launch has none.
- It is now written to disk the moment it is made and survives a restart. For a recording made by an earlier version the setting applies.
- A waiting recording now says in advance what will happen to its sound.
0.13.0October 30, 2026
An interrupted transcript carries on
- Transcribing an hour-long call takes minutes, and an interruption halfway through (quitting the app, a restart, a flat battery) sent it back to the start. Finished pieces are now kept as the work goes, and the next run continues from where it stopped.
- It carries on only under the same model, the same language and the same cut of the recording. If any of those changed, the recording is read again: a few extra minutes beat stale text in a transcript.
- The progress sits next to the recording and goes away with the finished transcript. A forgotten one is swept after two weeks.
0.12.1October 30, 2026
Long calls reach the end
- On recordings over an hour the transcript sometimes ended with the app quitting, and the work had to start over. Pieces were sent to the model several at a time, and they share its decoder state, which corrupted memory.
- Pieces now go one at a time. A long call takes longer than 0.10.2 promised, but it reaches the end.
0.12.0October 30, 2026
A note on a recording
- A recording can be signed in your own words. The first lines of the note show in the list card, where there used to be nothing but a date.
- The note is kept in the transcript file itself, beside the date and the model, so it travels with the folder and opens in any editor.
- The audio switch in settings now says what each position means. Next to a switch already turned on, a line about what happens by default read as a description of the current state.
0.11.0October 30, 2026
Your own term list
- A list of English words that must be spelled right: Cloudflare, GitHub, Core ML. It is given to the model before transcription and fixes what still came back spelled by ear.
- The match is on sound, so every mishearing and every ending lands on the same term, without an entry for each form.
- A term can be added straight from the transcript, and the list lives in settings next to the personal dictionary.
0.10.2October 30, 2026
One language per recording, faster transcripts
- In Auto mode the language is decided once from the start of the recording instead of inside every window, so a transcript no longer drifts into a neighbouring language halfway through.
- A long recording is transcribed noticeably faster: pieces go to the model several at a time, as many as the machine has memory for, and none at all in low power mode.
0.10.1October 28, 2026
A call is heard where it is read
- Kept call audio plays inside the app. Press the time of a reply and the recording starts there, while the text follows along and highlights the line being spoken.
- The recording strip got smaller: instead of a label for each side there is an icon, a colour and a live wave, so a dead track shows up at once.
- The transcript header was rebuilt: the title is no longer truncated, and length, replies and audio size sit on one line.
0.10.0October 27, 2026
The sound of a call can be kept
- A Keep audio switch right in the recording strip, which can be pressed at any point during the call, including its last second.
- Both tracks are mixed into one compressed file next to the transcript: an hour takes about 14 MB instead of 230 MB of raw tracks.
- Settings show how much space kept recordings take and can delete them all at once. By default the sound is still deleted as soon as the text exists.
0.9.1October 21, 2026
Menu items open the window again
- The Conversations and File transcripts items did nothing while the window was closed. Now any item opens the window itself and goes straight to that screen.
0.9.0October 21, 2026
An imported recording reads as text
- A transcript of a recording you bring in no longer carries timecodes: it is plain paragraphs. Calls keep their time, where it matters.
- When several voices are heard, the app tries to tell them apart and labels them Speaker 1, Speaker 2. It can be turned off in settings.
0.8.0October 21, 2026
One screen per thing
- Dictation, calls and file transcripts moved to screens of their own instead of sharing one list.
- The important fix: the text of transcribed files is finally visible in the app. The parser expected a speaker label on every line, and imported files have none.
0.7.1October 21, 2026
Folders do as they are told
- Folder buttons open the folder itself instead of revealing its icon in the parent one.
- The inbox folder can be any folder. Point it at the one where iCloud leaves Voice Memos from an iPhone and transcripts appear by themselves.
0.7.0October 21, 2026
Transcribe recordings you already have
- A handheld recorder, a phone, a voice message, the sound of a video. Drop the file on the app or leave it in the Inbox folder and it is picked up.
- Recognition runs on your Mac, and imported files are never deleted.
0.6.0October 15, 2026
The previous dictation is no longer lost
- The Dictation screen can keep a list of recent phrases, and a click puts one back on the clipboard.
- Nothing is kept by default. Settings turn it on for 5, 10 or 20 phrases, and they stay on your Mac.
0.5.1October 15, 2026
A long dictation no longer vanishes
- A dictation over ten minutes used to end with an error and the recording was deleted. Now it is cut at the pauses, recognised in parts and joined back together.
0.5.0October 14, 2026
You can see the recording happen
- A strip appears during a call: the time, the level of each side and the seconds already written to disk.
- Tracks are written into your own conversations folder, under Unfinished. If the app is closed mid call, the recording stays where it is.
0.4.0October 3, 2026
Dictation and call transcripts
- Press Record a call at the start and stop at the end. Your microphone and the sound your Mac plays are written as two tracks and merged into one Markdown file with timecodes.
- A Conversations screen appeared: recordings by time on the left, the dialogue on the right.
0.3.0October 2, 2026
Free local dictation
- Updates from inside the app: a Check for updates item in the menu bar.
- The build is signed with the project certificate for the first time, so macOS permissions survive updates.