Crate anda_db_hnsw 

Anda-DB HNSW Vector Search Library

A high-performance implementation of Hierarchical Navigable Small World (HNSW) algorithm for approximate nearest neighbor search in high-dimensional spaces.

HNSW is a graph-based indexing algorithm that creates a multi-layered structure to enable fast and accurate nearest neighbor search in high-dimensional spaces.

Features

• Fast approximate nearest neighbor search;
• Multiple distance metrics (Euclidean, Cosine, Inner Product, Manhattan);
• Configurable index parameters;
• Thread-safe implementation with concurrent read/write operations;
• Serialization and deserialization support;
• Support for bf16 (brain floating point) vector storage for memory efficiency.

Re-exports

pub use half;

Structs

HnswConfig

Tunable HNSW parameters. Defaults are suitable for 384–768-dim sentence embeddings; see the crate-level docs for guidance on tuning.

HnswIndex

Concurrent, persistable HNSW index for approximate nearest-neighbor search.

HnswMetadata

Index metadata.

HnswNode

One node of the HNSW graph.

HnswStats

Runtime statistics exported alongside the metadata.

LayerGen

Random layer generator for HNSW.

Enums

DistanceMetric

Distance metric used for similarity computation.

HnswError

Errors that can occur when working with HNSW index.

SelectNeighborsStrategy

Neighbor selection strategies used both during graph construction and when pruning over-connected nodes.

Functions

serialize_node

Serializes a node to CBOR. Used by HnswIndex::store_dirty_nodes and by external tools that snapshot individual nodes.

Type Aliases

BoxError