turbovec introduces Rust-based local vector indexing with Python bindings

turbovec is a new Rust-based vector index featuring Python bindings, designed for local, privacy-sensitive semantic retrieval. Built on Google Research’s TurboQuant algorithm, it utilizes a data-oblivious quantizer that requires no separate training phase or parameter tuning, allowing for online ingestion as a corpus grows.

The tool is optimized for memory efficiency and speed, claiming to fit a 10 million document corpus in 4 GB of RAM compared to 31 GB using float32. It utilizes hand-written SIMD kernels for ARM and x86 architectures to achieve search speeds that reportedly outperform FAISS IndexPQFastScan.

Key features include incremental saves that persist only changes since the last sync, crash-safe operations, and search-time filtering via ID allowlists. It is positioned as a solution for building air-gapped Retrieval-Augmented Generation (RAG) stacks where memory, latency, and data privacy are critical concerns.