Think about what your company writes down on purpose. Strategy decks get reviewed before they circulate. Contracts get read twice. Board minutes are drafted, edited, and approved. Now think about what your company says out loud: the salary someone is unhappy with, the customer who is about to churn, the diagnosis a patient just heard, the number a founder would never put in a slide. None of that is edited. All of it is spoken in meetings.
Meeting audio is the least filtered and most revealing material an organisation produces. It is the only record that captures hesitation, disagreement, and the thing someone said before they thought better of it. We treat it casually because for most of working history it evaporated the moment the call ended.
Transcription changed the physics
Once a conversation becomes text, it stops behaving like a conversation and starts behaving like a document. It can be searched, quoted, exported, indexed, subpoenaed, and pasted into a message thread by someone who was not in the room. That is genuinely useful — it is why we built a meeting recorder. But it means the storage decision matters far more than it used to. A transcript is not a recording you might listen back to. It is a permanent, greppable artefact of every unguarded sentence.
What a cloud notetaker actually asks of you
The standard offer is simple: connect your calendar, let a participant named "Notetaker" join every call, and get summaries in your inbox. In exchange, your organisation's spoken record is uploaded to infrastructure you do not operate, retained under a policy you did not write, and processed by systems whose behaviour can change with a release note.
This is not a claim that any particular vendor is careless. It is a structural observation. Once the audio leaves your machine, the questions that matter are no longer technical, they are contractual. How long is it kept? Which sub-processors touch it? Which jurisdiction is it stored in? Is it used to improve models, and if the answer is no today, what happens after the next funding round or acquisition? You can get answers to all of those questions. You cannot enforce them yourself, and you cannot verify them.
There is also the quieter cost: a third party now has a copy of your data, which means your risk surface includes their breach, their misconfigured bucket, and their employee access controls. You inherited a dependency you did not build.
The bot in the corner
Cloud notetakers usually announce themselves by joining the call. That is honest, and it is also a problem. A visible recording bot changes the meeting. Candidates get careful. Clients ask who else will see this. Colleagues stop saying the useful, half-formed thing. The tool meant to capture the conversation has quietly narrowed it.
The local alternative is not a compromise
The interesting development of the last few years is that you no longer need a data centre to transcribe speech well. Speech models now run acceptably on the laptop already sitting on your desk. The same machine that plays the meeting audio can capture it, write it to disk, and turn it into text without a single outbound request.
That is the architecture JesRecap uses. Your microphone and the meeting audio are recorded as two separate local files. Transcription happens on device with a model you download once. Speaker labels are computed locally. Nothing is uploaded — not the audio, not the transcript, not metadata about who you met. The only network traffic is the one-time model download and license activation.
We deliberately did not build an AI summariser into the app. Instead, meeting actions copy a ready-made prompt together with your transcript, and you paste it into whichever assistant you already trust. If that is a self-hosted model, nothing leaves your network. If it is a commercial one, that is your decision, made once per meeting, explicitly — not a default set by a vendor's pipeline.
The test we hold ourselves to
A privacy claim is only worth as much as its failure mode. If your internet connection dies mid-meeting, a cloud notetaker degrades. JesRecap keeps recording, because the recording was never going anywhere. If our company disappeared tomorrow, your meetings would still be on your disk, in ordinary WAV files and a local database, readable without us.
That is the whole argument. Meeting data is the most sensitive thing you produce, transcription makes it permanent, and the technology to keep it local is now good enough that sending it away is a choice rather than a necessity. We think it is the wrong choice, so we built the other one.