turbovec
In-process compressed vector index using TurboQuant quantization
What it is
turbovec is a vector search library for Rust and Python that compresses embeddings to 2 or 4 bits for each dimension and runs SIMD kernels over the compressed data without decompressing it. It needs no training step, supports adding and removing vectors instantly, incremental crash-safe saves to disk, and id allowlist filtering. Developers building retrieval or RAG pipelines can use it as a drop-in in-memory store for LangChain, LlamaIndex, Haystack, or Agno, keeping large embedding sets in a fraction of the RAM.