GitHub - open-puffer: opensource and fastest vectorDB
opensource and fastest vectorDB. Contribute to harishsg993010/open-puffer development by creating an account on GitHub.
Software Library
Summary
turbovec is a high-performance Rust-based vector index with Python bindings that utilizes Google's TurboQuant algorithm for zero-training, data-oblivious quantization. It offers significant memory compression and faster search speeds than FAISS, making it ideal for privacy-focused, local RAG applications.
Highlights
- Extreme memory efficiency, reducing a 31 GB float32 corpus to just 4 GB.
- High-speed search performance via hand-written SIMD kernels (AVX-512 and NEON).
- Zero-training quantization that requires no codebook training or data-dependent calibration.
- Efficient filtered search using allowlists that prevents over-fetching and maintains recall.
- Supports stable external IDs and O(1) deletions via the IdMapIndex.
auto-generated
Context
Audience
Machine Learning Engineers and Software Developers building high-performance, privacy-centric RAG applications
DomainMachine Learning
Formatopen source library
Accessopen source
Source Overview
turbovec is a Rust-based vector index with Python bindings built on Google’s TurboQuant algorithm, enabling zero-training, data-oblivious quantization.
It compresses large embedding corpora to a fraction of the RAM of float32 while matching or beating FAISS PQ performance, supports stable external ids, deletions, and efficient filtered search via SIMD-aware allowlists. The library integrates as a drop-in replacement for in-memory stores in LangChain, LlamaIndex, Haystack, and Agno, and is available on both PyPI and crates.io.
Topics
Related
Retrieval Augmented GenerationVector DatabasesFAISSTurboQuantLangChainLlamaIndex
