turbovec delivers training-free vector compression
turbovec is an open-source vector indexing library written in Rust with Python bindings that implements Google Research's training-free TurboQuant algorithm. The library compresses high-dimensional embeddings by up to 16x, utilizes hand-written SIMD kernels for fast CPU-based search, and integrates with LangChain, LlamaIndex, and Haystack.
turbovec solves one of the biggest bottlenecks of local RAG systems—memory consumption—by implementing a cutting-edge, training-free quantization algorithm. By avoiding the typical PQ training cycle, it enables true dynamic ingestion of vector embeddings, which is critical for real-time applications.
* Implements Google's data-oblivious TurboQuant algorithm to bypass the need for codebook training or training datasets.
* Delivers up to 16x memory compression, enabling developers to run massive vector datasets on standard consumer hardware.
* Hand-written SIMD kernels (AVX-512BW and ARM NEON) provide high CPU search performance that can surpass FAISS.
* Integrates directly into AI ecosystems with official bindings/wrappers for LangChain, LlamaIndex, and Haystack.
DISCOVERED
1h ago
2026-06-08
PUBLISHED
1h ago
2026-06-08
RELEVANCE