Where your words live

Nothing you say or type leaves this device unless you deliberately turn on sign-in or tick the optional transcription box. Audio never leaves, in any configuration. The analysis is a word list running in this browser — no model is sent your words.

A recording is written to this browser's IndexedDB and encrypted with AES-GCM before it is stored. The key is generated on this device and kept here. A copy of the database without the key is not readable text.

We never receive audio. There is no shared bucket of recordings, which means no-one at MIRRORSU can play yours — not as a policy we promise to follow, but because the file is not somewhere we can reach.

Transcripts, notes, themes, check-ins and skill answers live in this browser's local storage. They are stored as readable text rather than encrypted, because anything on this device that can be shown back to you can also be read by anything else with access to this browser profile. Treat an unlocked, shared device as shared.

MIRRORSU matches your words against a fixed list of words that ship inside the app, and counts what it finds. That is the whole mechanism. It runs here, it is the same every time for the same words, and it never involves a request to a server. It is also why nothing it tells you is a confidence score, a prediction, or a clinical measure — a word count cannot be any of those.

Sign-in is configured on this copy. It is still optional. signing in copies reflection metadata — the date, the themes, and the transcript you already have — to a database row scoped to your account, so another phone can show the same timeline. Audio is not part of that payload.

Those rows are protected by row-level security, so one account cannot read another's. They are not end-to-end encrypted: the server stores your written words in a form it can read. If that matters to you, do not sign in — every feature works without it, and a backup you export from Profile moves your data between devices without a server in the middle.

Turning speech into text happens in the browser, using the speech engine your browser ships. That engine may send audio to the browser vendor under their rules, not ours. In the App Store and Play Store apps it is the phone's own dictation instead — Apple's or Google's.

Server-side transcription is not enabled on this copy, so audio is not sent to a transcription vendor even if you tick the optional consent box. Typing never involves a speech engine at all.

An on-device activity log of which screens opened and whether a save succeeded. If optional telemetry is enabled, a beacon carries an area name, an event name, whether it worked, and how long it took. Never a transcript, never audio, never a word you wrote. There is no advertising analytics in this app.

From Profile you can export everything as a file, restore from one, or erase this device — which drops the encrypted audio database and the device key along with it. On a single reflection you can delete the recording and keep the words, or delete both.

Clearing your browser data, using private mode, or opening MIRRORSU on another phone gives you a new, empty copy. That is the trade for not requiring an account, and an exported backup is how you move between devices deliberately.

The privacy policy is the complete version, including retention, permissions and account deletion. If a sentence on this page disagrees with the app, the app is wrong and should be fixed.