Open Puffer is a single-node, NVMe-backed Rust vector database using segment-based storage with IVF-Flat, HNSW, IVF-HNSW, and PQ+OPQ for high-performance approximate nearest neighbor search.

It provides BM25-based full-text search, hybrid vector+text fusion (weighted, RRF, normalized, softmax), HTTP JSON APIs, and a CLI for managing collections, loading data, and running searches. The system supports metadata payloads, automatic compaction with tiered segments, mmap-based prefetching, and demonstrates sub-millisecond to low-millisecond latencies at million-vector scale with tunable recall via nprobe settings.