Note Indexing And Vector Database Persistence

via “segment-based storage with incremental updates”

A lightweight, lightning-fast, in-process vector database

Unique: Implements log-structured merge (LSM) tree principles for vector indexes, where new vectors are buffered in memory and periodically flushed to immutable segments, enabling efficient incremental updates without the full index rebuild overhead of traditional HNSW implementations

vs others: More efficient than rebuilding the entire HNSW index on each update (as required by pure in-memory systems), while simpler than Milvus's segment management because it avoids distributed consensus and uses local filesystem for persistence