Show HN: Homomorphically Encrypted Vector Database (github.com)

A new open-source project, HEVEC, has been unveiled, featuring a homomorphically encrypted vector database that enables real-time similarity search on encrypted data. This breakthrough means that both user data and queries remain encrypted throughout the client-server interaction, preventing any exposure of plaintext information to the server. HEVEC is designed to support large-scale applications, successfully demonstrating performance with approximately one million encrypted vectors processed in just 187 milliseconds. The significance of HEVEC lies in its potential to address critical privacy concerns faced by modern AI systems, particularly those that utilize personal embeddings, such as digital assistants and memory systems. Traditional vector databases necessitate plaintext embeddings for functionality, creating privacy bottlenecks. By maintaining encryption across all operations, HEVEC allows personal AI agents to function without revealing sensitive user data. The cryptographic foundation of HEVEC is based on the Module Learning With Errors (MLWE) scheme, which is anticipated to be secure against quantum attacks, emphasizing its robust security framework for future-oriented AI applications.