Meeting notes from audio, without sending the audio anywhere

Drop a recording and get a transcript, a summary and action items. The transcription runs on your device — a private meeting is not uploaded to anybody's server.

Drop an audio file here, or click to choose

mp3, wav, m4a, ogg, webm · transcribed locally · no upload

Transcript
Notes

Why the audio never leaves your machine — and why that is rare

A meeting recording is the most sensitive thing most people handle in a week: salaries discussed, decisions made, clients named, sometimes personal things said. Every mainstream transcription service works by uploading that recording to a server. "We delete it afterwards" is a promise about a copy that has already crossed the network.

This page runs Whisper inside your browser using WebAssembly. The audio is decoded locally, fed to the model in your tab, and the transcript is written back to your disk. There is no upload, because there is nothing to upload to. You can verify it the same way as everything on this site: open the Network tab, transcribe something, and watch no request carry your audio.

Choosing a model

Five models ship, and the right one depends on the job. Tiny is fast enough to feel instant and accurate enough for clear speech; start there. Base handles accents and overlapping voices noticeably better. Small is the most accurate and the slowest, and it is the right choice for a formal meeting where a name or a number must be right. Multilingual variants transcribe non-English speech; the English-only variants are faster and more accurate on English audio.

The first run downloads the model — 40 MB for Tiny, 250 MB for Small — and your browser caches it. After that it loads instantly, offline included. New and better models appear regularly; the list is a dropdown, not a decision, and a better model can be swapped in without changing anything else.

What the notes contain

The notes pane is built from the transcript itself: a one-paragraph summary, the decisions stated as decisions, the action items with owners where the transcript names them, and anything flagged as unclear. It is a starting draft for the notes you send — not a replacement for reading the transcript once.

Meeting notes questions, answered

Does my audio get uploaded?

No. The recording is transcribed on your device — nothing is sent to any server. The privacy test in the repo asserts no network request occurs while transcribing.

What formats can I import?

Common audio formats (m4a, mp3, wav, webm) work. The file is decoded locally and transcribed by an on-device model.

What does the output look like?

A structured markdown note: summary, key decisions, action items with owners, and open questions — ready to paste into your notes tool.

How accurate is the transcription?

It depends on audio quality and accents; crosstalk degrades accuracy. The summary and action items are extractive (drawn from the actual transcript), so they never invent a decision that was not made.

Can I edit the notes?

Yes — the output is plain markdown. Edit it in the tool or copy it anywhere.

Meeting notes questions, answered

Does my audio get uploaded?

No. The recording is transcribed on your device — nothing is sent to any server. The privacy test in the repo asserts no network request occurs while transcribing.

What formats can I import?

Common audio formats (m4a, mp3, wav, webm) work. The file is decoded locally and transcribed by an on-device model.

What does the output look like?

A structured markdown note: summary, key decisions, action items with owners, and open questions — ready to paste into your notes tool.

How accurate is the transcription?

It depends on audio quality and accents; crosstalk degrades accuracy. The summary and action items are extractive (drawn from the actual transcript), so they never invent a decision that was not made.

Can I edit the notes?

Yes — the output is plain markdown. Edit it in the tool or copy it anywhere.

How long can a recording be?

Practically, up to about an hour of clear speech on a modern laptop. The transcript and its summaries are built in chunks, so a long meeting takes proportionally longer but works the same way.

Can I use it for lectures or interviews, not just meetings?

Yes. Any spoken recording works — the action-item extraction simply finds fewer commitments in an interview, which is correct rather than wrong.