Privacy
My Plaud has no server of its own, no account and no telemetry. It contacts Plaud, and the providers you chose and gave a key for. This page lists what each of them receives.
At a glance
| Who | Receives | When |
|---|---|---|
| Plaud | Your login, and requests for the list of recordings and their audio | Every sync |
| The speech-to-text provider (ElevenLabs, OpenAI or OpenRouter) | The audio file, the model, the language, and key terms where used | Once per recording, and when you transcribe again |
| The cleanup model’s provider | The text of each turn, without speaker labels | Once per transcript |
| The summary model’s provider | The transcript with speaker labels | Once per transcript, and when you regenerate a summary |
| Hugging Face | A download request for the Whisper model | Only when you press Download |
Never sent to a speech-to-text or model provider: Plaud’s title, the folder name, the recording ID or account data. The video of a recording made on the Mac is sent to no one.
Plaud
The app only reads from Plaud. It never changes or deletes anything in the Plaud cloud. A recording made on the Mac is never sent to Plaud.
Speech to text
ElevenLabs: the audio file, under the name audio.ogg or audio.mp3, with the language, the model and the key terms you configured. ElevenLabs keeps uploaded audio and transcripts by default. With “Delete the transcript at ElevenLabs after saving” on (the default), the app deletes each transcript from ElevenLabs as soon as it is stored locally. Whether that also deletes the uploaded audio is not documented; check the ElevenLabs dashboard if it matters to you.
OpenAI and OpenRouter: the audio file under the same name, the model and the language; to OpenRouter also the key terms. With OpenRouter the audio goes to OpenRouter and to the provider that serves the chosen model.
Local: nothing. The recording is transcribed on the Mac.
Cleanup and summaries
- Cleanup sends fixed instructions and the numbered turn texts, without speaker labels.
- Summary sends fixed instructions and the transcript as
transcript.mdshows it, speaker labels included, with your corrections and the names you entered.
Only the provider of the chosen model receives the text. Requests to OpenAI ask it not to store the response. With OpenRouter the text also reaches the upstream provider that serves the model; the requests ask OpenRouter to route only to providers that do not keep or train on prompts. What each provider keeps beyond that is set by its own terms.
With Local, the text goes to the model server on the Mac and nowhere else, unless you point Local server at another machine.
Keeping everything on the Mac
Choose Local as the speech-to-text provider and Local for the cleanup and summary models. The only network contacts are then the downloads from Plaud and the one-time download of the speech model.
On the Mac
- The login tokens and the API keys are stored in files readable by your user only.
- The keys are never written to the settings file or to the log.
- The log holds recording IDs, HTTP statuses, error codes and counts. Recording names, transcript text, summaries, provider error messages and the account’s email and name are never written to it.
Recording other people
Recording someone without telling them may be against the law where you or they are. Tell the people you record.