Releases

Available as a Developer ID-signed, Apple-notarised Mac download for Apple Silicon on macOS 14.2 or newer, and on Microsoft Store for Windows 10 and 11. An unsigned Windows x64 installer is also available as a secondary download.

• Recording recovery repairs both audio tracks independently and preserves damaged audio for recovery.
• Model downloads resume after interruption, check their checksum and respond to cancellation when a connection stalls.
• Revised prompts keep short summaries proportional to their transcripts. Failed or incomplete AI generations preserve previously saved notes.
• Oversized local-note requests explain the model limit and let you copy the complete prompt for another model.
• Settings panels restore their selection when the app regains focus.

AI transcripts and notes can still get names, numbers and decisions wrong. Check important details against the recording. These downloads use the regular website license; Apple purchase testing is a separate build.

Available from the downloads page for macOS and Windows. What changed:

• Speaker labels no longer hold up the transcript. Transcription finishes and saves on its own, and labelling runs afterwards as a background job you can start per meeting, with a time estimate measured on your own machine rather than a guess
• Labelling never overwrites a speaker name you typed yourself, and an interrupted run is offered as a restart instead of vanishing
• Whisper models are gone. Transcription is Parakeet (25 European languages) or Moonshine (English, lightest), both on one runtime, for a smaller install and less CPU, which matters most on Windows
• AI notes now uses a 530 MB model instead of 1.3 GB, and moved out of the launcher into Settings where it belongs as an opt-in
• Prompt buttons keep a copy-to-clipboard escape hatch even with AI notes on, so a weak local note is never a dead end
• Custom prompts: as many as you like, each named, reorderable and deletable, each with its own button and its own note

The previous preview for both platforms. What it included:

• Searchable meeting library, stored locally in SQLite
• Dual-track capture: microphone and system audio as separate files, plus a mixed playback track
• On-device transcription with local speech models, with an optional live transcript during the meeting
• Speaker labels applied after recording by a local diarization model
• Meeting actions that copy a summary, action-items or custom prompt with the transcript included
• Exports: Markdown, plain text, JSON archive, SRT and VTT captions, and WAV audio
• Optional AI notes written on your device by a local model (off by default)
• Import of existing audio files: WAV, MP3, M4A, AAC, FLAC, OGG, WebM
• A choice of where new recordings are stored

Earlier builds in the 0.1.x series were narrower steps towards the current one, and we are not going to reconstruct changelogs for them after the fact. If you are on an older preview, the fix is the same either way: install the current build from the downloads page. Your existing meetings are untouched.