Why Your Meeting Audio Should Never Leave Your Device
A meeting recording is one of the most sensitive files a person handles: salaries discussed, decisions made, clients named, sometimes personal things said. Most transcription services work by uploading that recording to a server and promising to delete it afterwards. This tool does not upload anything — Whisper runs inside your browser, on your device, and the audio never crosses the network.
What a cloud transcription service sees
With a cloud service, the recording leaves your machine and lands on a provider's server. Even when the provider says they delete it, the file has already crossed the network. It travelled across cables and routers you do not control, sat on someone else's disk, and may have been logged by a load balancer, a CDN or a backup job before it was transcribed. You are trusting the provider's security, their staff, their retention policy and their breach response — all for a copy you never asked them to make.
What local transcription sees
Running Whisper in the browser means the model is downloaded once, cached by your browser, and run with WebAssembly inside this tab. The audio file is decoded locally, fed to the model in memory, and the transcript is written back to your disk. There is no server to upload to, so nothing is uploaded. You can verify it yourself: open the Network tab in Developer Tools, transcribe a recording, and watch no request carry your audio.
Why that matters for meetings
Meetings cover pricing strategy, personnel issues, client names and roadmap decisions. A transcript that never leaves the device is a transcript that cannot be intercepted in transit, scraped from a provider's logs, or used to train someone else's model. Local processing turns privacy from a promise into a property of the tool — it works the same on a café's Wi-Fi as it does on your office network, because nothing crosses the wire.
The trade-off
The model has to download before the first transcription — 40 MB for the smallest model, more for larger ones — and it uses your device's CPU or GPU while it runs. That is the cost of keeping the audio local: you pay it once, in time and battery, instead of paying it in data you no longer control. For a private meeting, that is usually the better deal.
The model download is the only network request
When the page loads, the transcription model itself is fetched from a content-delivery network and cached by your browser. That download is the one request the site makes that carries data from a remote server to you. After that, the model lives in your browser's cache and runs offline — you can switch your Wi-Fi off and transcribe a recording with no connection at all. The audio itself never appears in any request.
Use it on a network you do not trust
Because nothing is transmitted, the tool works the same on a café's Wi-Fi, an airport network or a hotel connection as it does on your office network. There is nothing to intercept. A tool that uploads cannot make that claim, and a VPN does not change it — an encrypted connection still delivers the recording to a server that stores a copy.
Verify it yourself
Do not take the privacy claim on trust. Open Developer Tools, go to the Network tab, and transcribe a recording. The only entries that appear are the page's own files and, if advertising is live, an ad request. No entry carries your audio. If you prefer a stronger test, disable the network entirely after the model has loaded and transcribe anyway — it still works.