Never uploaded to be read
The audio bytes, the unreleased master, anything a model could train on as music.
Proof
Your track is read in this browser and never uploaded to be read, so we cannot train on your music. That is architecture, not policy.
Open thermal-ai.app, then turn on airplane mode.
Drop the file on the homepage instrument. Tempo, key, and the rest of the read print in this browser.
Once the page has loaded, the worker is here. The decode never waits on a server.
Offline before the page loaded? Reload once you are back online, then repeat.
| Request | Carries the audio | Needed for the read |
|---|---|---|
| None to Thermal | No | No. Decode and measurement stay in the page. |
| Analytics ping, if you are online | No | No. Path and event name only. Not the file, not the fingerprint. |
The audio bytes, the unreleased master, anything a model could train on as music.
A compact measurement object, if you start a written kit, a mix read, sonic neighbors, or a public page: tempo, key, loudness as dBFS, word labels, and a short energy envelope. Not the file.
Your track itself leaves only when you send it: an upload to your own SoundCloud, or a video you post to an account you connected. A post's file is held privately until it is sent, then deleted.
The distinctiveness score stays in the studio: never on a public page, never how the directory is ordered.