Privai Sandy

Privai Sandy is a proof-of-concept privacy layer for frontier AI models. It aims to filter sensitive information, such as tax documents, health data, and financial transactions, from prompts before they reach the model.

The creator notes that catching all sensitive information while keeping response quality intact is difficult, so this is an early experiment rather than a finished product. Venice AI, which limits identity connected to prompts but not the information within them, was cited as a similar approach. The source code can be shared on request.