The short version: your prompts go to the model you picked and nowhere else, and we do not train on anything you send.
An account identifier, your balance, your API keys as hashes, and a usage record of which model was called, when, and what it cost. That record is what makes the account page and your spend history possible.
A request has to reach the model to be answered, so the prompt is passed to whichever provider you selected and is handled under their terms for that call. Choosing a different model changes who sees it.
Conversations can be deleted from the account at any time. Usage records are kept in aggregate for accounting, without prompt content attached.
Questions about any of this: ask@kytro.xyz.