nesaorg/nesa

Leverages **Equivariant Encryption (EE)**, a novel cryptographic approach that enables neural network inference directly on encrypted data without the latency penalties of homomorphic encryption—achieving near-zero overhead by exploiting model architecture properties rather than relying solely on arithmetic operations. Supports a broad model ecosystem including Llama, Mistral, and Stable Diffusion, with a ChatGPT-compatible API enabling one-line integration for encrypted inference workloads.

3,070 stars. No commits in the last 6 months.