Meeting transcription that stays in the room

Meeting audio is the most sensitive audio there is: strategy, salaries, customers, disagreements. Sending that to a cloud transcription service — or inviting a bot to sit in the call — means trusting someone else’s retention policy with your most candid conversations.

This tool takes the opposite approach: the recording is transcribed in your browser, on your machine, and nothing else ever sees it. Your files are never uploaded — everything runs on your device. No meeting bot, no third-party processor, no account to breach.

Checking what this browser supports…

How to transcribe a meeting

  1. Export the recording

    Download the audio or video from your meeting platform (or use a local recording). MP3, M4A, WAV, and MP4 all work directly.

  2. Drop it in — locally

    The transcript streams in with timestamps while the file processes. An hour-long meeting keeps its progress and ETA visible throughout.

  3. Search, correct, share deliberately

    Find the decisions with the search box, fix names inline, and export TXT or JSON. What leaves your machine is a file you chose to share.

For confidential conversations

No third-party processor

Cloud transcription adds a company to your meeting’s audience. Local transcription adds no one — the audio is processed and discarded from memory on your own device.

Simpler compliance story

Explaining a tool that never transmits audio is easier than reviewing another vendor’s data-processing agreement. Your recording never crosses the network at all.

Long meetings welcome

No minute caps means the two-hour all-hands is as transcribable as the five-minute stand-up. Progress streams the whole way.

Meeting transcription questions

Is this safe for confidential meetings?
The architecture is the answer: the recording is never uploaded, so no third party — including us — can access it. For regulated environments, that also means one less vendor in your data-flow diagram. (This is a description of how the tool works, not legal advice.)
Does it label who is speaking?
Yes, for up to 8 people per recording. Switch on Identify speakers and each line gets a speaker label you can rename; the names carry into every export. It runs NVIDIA's Nemotron 3 Diarization model in your browser, a 120 MB download the first time.
How do I get the recording out of Zoom / Teams / Meet?
All three offer a recording download (typically MP4 or M4A) to the host after the call. Drop that file here as-is — the audio track is extracted automatically.
Can it handle a two-hour meeting?
Yes. Long recordings are processed in windows with streaming results, so memory stays bounded and you can read the first half while the second half transcribes.
What does the transcription locally?
OpenAI’s open-source Whisper model, executed by our open-source engine inside your browser — WebGPU when available, WebAssembly otherwise.
Do you keep any copy of the transcript?
No. The transcript exists in your browser tab and in whatever exports you download. Close the tab and it is gone from everywhere, because it never existed anywhere else.
How can I check that my recording is not being uploaded?
Open your browser's developer tools, switch to the Network tab, and start a transcription with it open. On the first run you will see the Whisper model download once — about 80 MB for the standard model, less or more if you pick a different one — and after that it is served from your browser's cache. What you will not see, on any run, is a request carrying your recording. Then sort that list by size: the largest thing the page sends is a few kilobytes of analytics, nowhere near the size of a meeting recording. This page loads analytics and advertising like most websites do, which is why the claim is scoped precisely: your files are never uploaded.