What this is actually for
Not every recording needs this. Here's when it's the right tool, and when it plainly isn't.
A recording that can't be uploaded
A client interview, a patient session, an HR investigation, a source who never agreed to have their voice sitting on someone else's server. If the recording itself is the problem, a cloud service is the wrong tool no matter how its policy reads — a policy is a promise about data somebody else is holding. This one never receives the recording in the first place, so there's nothing to promise.
A locked-down work laptop
No admin rights, no App Store, no exceptions — the usual advice, "just run Whisper locally," isn't available to you. Opening a web page isn't a privileged action, so this runs anyway.
If you need to check with IT first, this is an accurate description to give them:
- A static web page — no installer, no executable, no extension, no elevated permissions.
- Audio is read from local disk and processed in the browser. Never transmitted, and neither is the transcript.
-
Ships
connect-src 'self'in its Content-Security-Policy, which is verifiable from the response headers, not just claimed. - About 235 MB of model files download once into the browser cache. Clearing site data removes everything.
- Open-source components, listed with licences at /licenses.txt.
Two things we won't claim: that this is a compliance certification, or that your organisation's policy permits it. That's still your call — this just gives you an accurate basis for asking.
The longer version of this, with a description written to be sent to IT — including what they tend to ask back, and the three cases where a browser tool won't rescue you.
No account, no meter running
Cloud transcription costs its provider money for every minute, so a free tier needs a floor — a cap, an account, an email worth marketing to. This has no per-minute cost to recover, because your own processor does the work, so there's nothing to meter and no login to create. If you're moving off Otter, Fireflies or Rev because the free minutes ran out or a bot joining the call became the problem, this is what that looks like instead.
Is it the right fit?
Use Ratproof if…
- the recording shouldn't be uploaded, full stop
- your laptop won't let you install anything
- you want a transcript to search and work from, checking quotes against the audio
- you'd rather see a measured error rate than a promise
Use something else if…
- you need speaker labels today — built, but not shipped yet
- you need to quote it without checking; hire a human
- you're on a phone, or a machine that can't spare a couple of GB
- the audio isn't sensitive and you need it back in minutes — a cloud service is faster
The cost either way
Your machine does the work a server would, so it takes longer and wants more memory. Every exact figure is on the measurements page.