LMDiskANN.jl: An Implementation of the Low Memory Disk Approximate Nearest Neighbors Search Algorithm

LMDiskANN.jl (v1.2.0) is a Julia package that implements the Low-Memory Disk Approximate Nearest-Neighbor (LM-DiskANN) algorithm, extending DiskANN-style graph search to handle billion-scale vector datasets while keeping RAM usage to a minimum. It stores adjacency lists on disk via memory-mapped files, performs tunable best-first graph traversals for fast and accurate queries, and supports dynamic insertions and deletions with automatic pruning to maintain a compact index. The library exposes knobs to balance recall against latency, and it optionally pairs a LevelDB key–value store with the node IDs for flexible external key lookup. These capabilities make LMDiskANN.jl well-suited for embedding retrieval, recommendation systems, and other large-scale similarity-search workloads that need high throughput on commodity hardware.