AlayaLite
A fast, lightweight vector database with a Python API, offering an in-memory HNSW-based index for low-latency search and an on-disk quantized graph index for billion-scale datasets that exceed RAM.
AlayaLite is a C++ vector database with a Python API, aimed at approximate nearest-neighbor search. Its in-memory path pairs a graph index with RaBitQ quantization for low-latency retrieval, and its Collection API returns ANN results together with document IDs, text and metadata filters for hybrid search.
For datasets that do not fit in RAM, the LASER on-disk quantized graph index keeps hot data on SSD and only the active working set in memory, using platform-specific backends: libaio on Linux, a thread pool on macOS and IOCP on Windows. It scales across threads using C++20 coroutines and installs via pip or uv, with LASER's extra dependencies such as scikit-learn and faiss-cpu as an optional install.