Ruvector Extensions

ruvector-extensions

Advanced feature pack for RuVector providing embedding generation pipelines, a built-in admin UI, data export to multiple formats, temporal version tracking, and pluggable persistence adapters.

Quick Reference

Installation

npx ruvector-extensions@latest

Quick Start

import {
  EmbeddingPipeline,
  startUI,
  Exporter,
  TemporalIndex,
} from 'ruvector-extensions';
import { HnswIndex } from 'ruvector-core';

// Embedding pipeline
const embedder = new EmbeddingPipeline({ model: 'all-MiniLM-L6-v2' });
const vectors = await embedder.embed(['Hello world', 'Vector search']);

// Create index and insert
const index = new HnswIndex({ dimensions: 384 });
index.insertBatch(vectors.map((v, i) => ({ id: `doc-${i}`, vector: v })));

// Start admin UI
await startUI({ port: 3000, index });

// Export to Parquet
const exporter = new Exporter();
await exporter.toParquet(index, './export.parquet');

Core API

EmbeddingPipeline

Generate embeddings from text using ONNX models.

const embedder = new EmbeddingPipeline(config: EmbeddingConfig);

EmbeddingConfig:

// Embed texts
const vectors = await embedder.embed(texts: string[]): Promise<Float32Array[]>

// Embed single text
const vector = await embedder.embedOne(text: string): Promise<Float32Array>

// Get model info
embedder.dimensions: number
embedder.modelName: string

startUI(options)

Launch a web-based admin dashboard for exploring and managing indexes.

await startUI(options: UIOptions): Promise<Server>

UIOptions:

Exporter

Export index data to common formats.

const exporter = new Exporter();

await exporter.toParquet(index, './out.parquet');
await exporter.toCSV(index, './out.csv');
await exporter.toJSON(index, './out.json');
await exporter.toNDJSON(index, './out.ndjson');

Export methods:

TemporalIndex

Wraps an index with version tracking and point-in-time queries.

const temporal = new TemporalIndex(index: HnswIndex, options?: TemporalOptions);

TemporalOptions:

// Insert creates a new version
temporal.insert('doc-1', vector, metadata);

// Query at a specific point in time
const results = temporal.searchAt(query, k, { timestamp: Date.now() - 86400000 });

// Get version history for a vector
const history = temporal.history('doc-1');

// List available snapshots
const snapshots = temporal.listSnapshots();

// Restore from snapshot
await temporal.restoreSnapshot(snapshotId);

Persistence Adapters

Pluggable persistence backends.

import { SqlitePersistence, S3Persistence } from 'ruvector-extensions';

// SQLite persistence
const sqlite = new SqlitePersistence('./vectors.db');
await index.save(sqlite);
const restored = await HnswIndex.load(sqlite);

// S3 persistence
const s3 = new S3Persistence({ bucket: 'my-vectors', prefix: 'indexes/' });
await index.save(s3);

References

• API Reference
• npm