What is fully-homomorphic-encryption?

This repository is the home of Google's Fully Homomorphic Encryption (FHE) effort, run by the FHE Team at Google in partnership with outside collaborators. Fully homomorphic encryption is a technique that lets a computer perform calculations on encrypted data without ever needing to decrypt it first. This matters because it means sensitive data, like medical records or financial information, could be processed by a third party or in the cloud without that party ever seeing the actual values. Only the person who holds the decryption key can read the results. The stated goals of the project are described simply as three aims: make FHE easy to use, make it fast, and make it work at real scale. What started here as a single C++ transpiler (a tool that converts ordinary code into an encrypted version) has since grown into two separate projects that are now maintained in their own repositories. HEIR is a compiler toolchain that takes existing models or programs and converts them into versions that can run under FHE, supporting multiple encryption schemes and target environments. Jaxite is a backend focused on running FHE computations efficiently on GPUs and TPUs (specialized chips used in machine learning), written using JAX, a Google library for numerical computing. This repository now mainly serves as a landing page pointing to those two successor projects rather than a place for active development itself. The original transpiler code is preserved in a tagged archive release for anyone who needs it, but new work happens in HEIR and Jaxite. Because the underlying subject is cryptography research, this project is best suited to engineers and researchers already comfortable with encryption concepts, rather than someone looking for a drop-in tool to add privacy to an app without any cryptographic background.