buildt-ssdb is a simple, efficient, and easy-to-use implementation of the Hierarchical Navigable Small World (HNSW) algorithm for approximate nearest neighbor search. It's perfect for searching mid-scale, high-dimensional datasets quickly and with minimal memory overhead.

This package has no dependencies, so should be usable both in the browser and Node scenarios.

To install buildt-ssdb, run the following command in your project directory:

npm install buildt-ssdb

npm i

followed by

npm run build

Here's an example of how to use buildt-ssdb:

import { HNSW } from 'buildt-ssdb';

// Create an HNSW index with the cosine similarity metric, 5 dimensions, M = 16, and ef = 50
const hnsw = new HNSW('cosine', 5, 16, 50);

// Add nodes to the index
const nodeId1 = 0;
const vector1 = new Float32Array([0.1, 0.2, 0.3, 0.4, 0.5]);
hnsw.addNode(nodeId1, vector1);

const nodeId2 = 1;
const vector2 = new Float32Array([0.5, 0.4, 0.3, 0.2, 0.1]);
hnsw.addNode(nodeId2, vector2);

// Perform a k-nearest neighbor search with k = 3
const queryVector = new Float32Array([0.15, 0.25, 0.35, 0.45, 0.55]);
const nearestNeighbors = hnsw.searchKNN(queryVector, 3);

console.log('Nearest neighbors:', nearestNeighbors);

The largest dataset I've pushed through this is a 100k vector index at 1536 dimensions per vector. The searchKNN method took ~2.36ms per query at this scale. I haven't gone any larger than this yet but likely will give it a try when I have the time.

The main class for working with HNSW indices.

Constructor

constructor(similarityMetric: 'cosine' | 'euclidean', numDimensions: number, M: number, ef: number)

addNode(id: number, vector: Float32Array): void: Add a node to the index. deleteNode(id: number): void: Delete a node from the index. searchKNN(queryVector: Float32Array, k: number): HNSWNode[]: Perform a k-nearest neighbor search. getNodeById(id: number): HNSWNode | undefined: Get a node by its ID. getSize(): number: Get the total size of the index (number of nodes). serialize(): Uint8Array: Serialize the index to a binary format. static deserialize(data: Uint8Array): HNSW: Deserialize an index from its binary representation.

I want to add capabilities for sharded indices, as that would be very useful for Buildt, as well as having a temporal component to the vector database – something I haven't yet seen from the main providers.