Features

Archive

  • Downloads the audio of every recording in one Plaud account, as the original file.
  • Syncs every 15 minutes (the interval is a setting), at launch and after the Mac wakes; Sync now does it at once.
  • One folder per recording, named by local date and time, in year/month folders. With a summary model, the generated name is added to the folder name.
  • Never changes anything in the Plaud cloud.
  • Nothing is deleted for good. A recording removed in Plaud stays in the archive, marked as removed; a setting can move it to the Trash instead.
  • Delete… moves a recording’s folder to the macOS Trash, and a sync does not download it again. Put it back from the Trash to undo.
  • Full check verifies every audio file against its checksum and downloads damaged ones again.

Transcription

  • Four providers, chosen in Settings: ElevenLabs Scribe, OpenAI, OpenRouter, or Whisper on this Mac.
  • A transcript is transcript.md next to the audio: a speaker and a time on every turn.
  • A language setting (or automatic), and with ElevenLabs and OpenRouter a list of key terms the model should know.
  • On this Mac: Whisper large-v3 turbo in Core ML, about 4 minutes for an hour of audio on an M4. The 1.7 GB model is downloaded only when you press Download.
  • Transcribe again… for the recordings you choose, with the hours of audio and the cost (or the time, on this Mac) shown before you confirm.
  • Failed recordings are retried on their own, three times, and then with Retry.
  • With ElevenLabs: the balance in hours on the Overview, a warning when it runs low, and the remote transcript deleted once it is stored locally.

Speakers

  • Speakers are told apart in every recording.
  • Name a speaker in the recording’s detail; the name shows in the transcript, the list and the summary.
  • A voice sample of up to 59 seconds is cut for each unknown speaker, ready to upload to the ElevenLabs speaker library.
  • A speaker in the library is recognised in later recordings, and one name serves all of them.
  • Select all with unidentified speakers picks the recordings worth transcribing again after you added someone.

Cleanup, names and summaries

  • A language model fixes punctuation, capitalisation, mis-recognitions and broken sentences, and removes filler words and false starts.
  • Speaker labels and times never pass through the model, so they cannot be changed by it.
  • A model writes a short name and a summary (summary.md) for every recording.
  • Providers: Anthropic, OpenAI, OpenRouter, or Local, a model server on this Mac such as LM Studio.
  • Cleanup and summary can use different models and different providers.
  • Regenerate summary… writes one recording’s summary again, from the transcript as you corrected it.
  • The raw transcript is kept, so nothing the model did is irreversible.

Recording on this Mac

  • Start recording in the menu records the microphone mixed with everything the Mac plays.
  • The same result with speakers or headphones, wired or Bluetooth, at any volume.
  • Three modes: audio, audio and screen, audio and one window.
  • Pause, Resume and Stop; the pause is not in the file.
  • Starts at once, also while a sync or a transcription runs.
  • Stored when the app quits, the Mac sleeps or the disk fills up; what was written before a crash is recovered at the next launch.
  • A recording made here is never sent to Plaud.
  • Audio is about 15 MB an hour (Ogg Opus), video about 700 MB an hour (HEVC).

The window

  • Overview: what needs attention, the last sync, the Plaud session, the ElevenLabs balance, the archive folder.
  • Recordings: every transcribed recording by date, name, duration, speakers and state.
  • Recording detail: a player with 15 seconds back and forward and four speeds; the transcript with the turn being played highlighted; the summary; the video, when there is one.
  • Correct the text of a turn, rename a recording, name a speaker. Corrections are stored apart from the transcript and can be reverted.
  • Settings: one form for the archive, the keys, transcription and cleanup.
  • Diagnostics: one line per check, from the session to the microphone permission.

The menu bar

  • One status line: up to date, what is running and how far it is, or how many things need you.
  • Warnings with the button that fixes each one: Log in, Retry, Add key…, Open System Settings.
  • A progress bar and Cancel while a job runs.
  • The icon shows a ring while working, an exclamation mark when something needs you, and a dot with the elapsed time while recording.
  • No Dock icon unless the window is open.

Privacy and files

  • Ordinary files in a folder you choose; no database, no lock-in.
  • Your own API keys, stored in files only you can read.
  • No My Plaud server, no account, no telemetry.
  • Recording names, transcripts, summaries and keys are never written to the log.
  • A fully local setup is possible: Whisper on this Mac and a local model.
  • Open source under the MIT licence.

Download · Documentation