How to tell if a meeting app really transcribes locally

Sean · 7 August 2026 · notes from an unfinished build

Every meeting recorder now says something about privacy. Most of them mean it, in the sense that they are not lying about anything specific. The trouble is that "private", "secure" and even "on-device" get used for arrangements that are quite different from each other, and the difference matters exactly when the meeting is sensitive enough for you to care.

There is a test that settles it, and it takes about ten minutes.

The test

Start a recording in the app. Say a few sentences. Then turn off your machine's network — Wi-Fi off, cable out, airplane mode, whatever is quickest — and keep talking for another minute. Turn the network back on and look at the transcript.

No marketing page survives this test, because it does not ask the vendor anything. It asks the software.

Why the label is not enough

The common arrangement that surprises people is a tool that keeps your notes local while sending the audio to a hosted speech API. The summary sits safely on your disk; the raw conversation — the part with the salary figure, the client name, the thing someone said before they thought better of it — has already left. If what you care about is the recording of the meeting rather than the file of bullet points, that setup gives you the opposite of what you wanted.

A second one worth knowing: some apps transcribe locally but send the transcript to a cloud model for the summary. That may be perfectly acceptable to you. It is simply a different question from the one the privacy page answers, so ask it separately.

Two follow-up questions

If the app passes the network test, two more things decide whether it will actually work for your meetings:

Does it capture your computer's audio, or only your microphone? A microphone-only recorder hears the people on your call as whatever leaks out of your speakers, processed through echo cancellation that is designed to remove exactly that. You will sound fine. Everyone else will not.

Are the two audio sources kept apart? If your microphone and the incoming call audio are mixed into one file before transcription, the software has to guess who was speaking. If they are kept as separate tracks, it does not have to guess at all in a two-person call, and a diarization model only has to sort out the far side.


I am building a local meeting recorder for reMarkable tablet owners, which is why I have opinions about this. It runs entirely on the Mac, and it passes the test above by construction — there is nowhere for the audio to go. There is no public installer yet. What it is and where it honestly stands.