What LAMBR does with your voice.

If you are not signed in, LAMBR writes nothing down. Your recording is sent once to be transcribed, the transcript is read for the shape of your reasoning, and the result is handed back to your browser. No row, no file, no account.

If you are signed in, the session is saved to your account: the problem, the transcript, the map made from it, and the timings. The audio itself is kept only if you tick a box, and can be deleted from the session afterwards.

Nothing is sold, nothing is advertised against, and no analytics service is running on this deployment.

Taking the challenge from the landing page without an account writes nothing to LAMBR's database or storage. The code is in app/api/transcribe/route.ts and app/api/analyze/route.ts: both persist only when there is a signed-in user, and there is no anonymous branch that writes.

Your recording still leaves your browser. It goes to LAMBR's server, and on to the transcription provider named below, because that is how a transcript gets made. Neither keeps it afterwards.

Your result is held in the browser tab itself, in sessionStorage, so that you can keep it if you decide to make an account. Closing the tab discards it. LAMBR cannot read it.

Nothing is held that is not named here. Your own copy of this list, with live counts, is on your account page.

These are the only companies involved. Each one receives what is listed and nothing else.

No analytics or product-measurement service is running on this deployment. The code for one exists and is inert without a key, and no key is set. If that ever changes, this page changes with it — it is the first thing here that would have to.

Nothing is sold or shared for advertising. There are no advertising trackers, no third-party pixels and no data brokers.

Nothing you said out loud is shown to anyone else. A comparison with a friend shows the two shapes and the counts, and the transcripts are not fetched to build it.

Population figures are aggregates with a floor: below five sessions LAMBR shows nothing, and withholds the count as well, so that a small group cannot be narrowed down by watching the number move.

It is asked as its own question, it is never bundled with anything else, and refusing changes nothing about what LAMBR does for you.

Saying yes means the session — including the transcript, because the words are the evidence — may be included in research on how people reason, including work with research partners. An export carries no email, no name and no account id: your sessions are linked to each other by a one-way salted hash and to nothing else.

You can withdraw it at any time from the session itself, and withdrawal is retroactive for every export made from then on. It cannot reach an export already produced and handed over, which is why the page says so rather than promising an undo.

LAMBR can be installed to a home screen, which means it has a service worker, which means it has a cache. That cache holds exactly one class of thing: the site's own build output — scripts, styles and icons that are identical for every visitor and contain nothing about anybody.

It deliberately does not cache pages. A signed-in page carries the transcript of you thinking out loud, and storing it for offline use would write your voice, as text, into a cache on that device — where it would survive signing out. So navigations are fetched from the network every time and never stored, and requests to LAMBR's own interfaces and to Supabase are not touched by the worker at all.

The consequence is honest rather than convenient: with no connection, the installed app shows a page saying so instead of your history. Uninstalling it, or clearing site data, removes the cache entirely.

Two small things are kept in ordinary browser storage: whether you dismissed the offer to install, and — only if you took a challenge without an account — that one result, so you can keep it if you sign up. The second is in sessionStorage and is gone when you close the tab.

Sessions are kept until you delete them. LAMBR is a record of how your reasoning moves over months, which does not work if it quietly expires — so nothing is deleted on a timer and nothing is deleted without you asking.

A stored recording is kept until you delete it or delete your account. Deleting a recording does not remove the transcript that was made from it, and the page that offers the deletion says so before you press it.

All of this is in the product rather than behind a request form.

Every read and write the website makes runs as you. The database decides what you can see through row-level security, rather than the application filtering results it was trusted to filter — so a bug in a page cannot show you somebody else's session, because the rows never arrive.

The recordings bucket is private and keyed to your user id. The website holds no key capable of bypassing any of this; the only component that does is the background worker, which is not reachable from the internet.