@ruvector/node: Unified RuVector package for Node.js — high-performance vector database with GNN capabilities.

# @ruvector/node [![npm](https://img.shields.io/npm/v/@ruvector/node)](https://www.npmjs.com/package/@ruvector/node) [![license](https://img.shields.io/npm/l/@ruvector/node)](https://github.com/ruvnet/ruvector/blob/main/LICENSE) One install for the native RuVector vector database and its graph-neural-network (GNN) layer in Node.js. This package re-exports [`@ruvector/core`](https://www.npmjs.com/package/@ruvector/core) (`VectorDB`) and [`@ruvector/gnn`](https://www.npmjs.com/package/@ruvector/gnn) (`RuvectorLayer`, `TensorCompress`, …); both are native N-API modules installed as dependencies. ## Install ```bash npm install @ruvector/node ``` Requires Node.js 18+ on a platform with prebuilt binaries for both dependencies (Linux x64/ARM64 glibc, macOS x64/ARM64, Windows x64). ## Usage ```javascript const { VectorDB, RuvectorLayer, TensorCompress, differentiableSearch } = require('@ruvector/node'); // Vector database (from @ruvector/core) const db = new VectorDB({ dimensions: 4, storagePath: './vectors.db', hnswConfig: {} }); await db.insert({ id: 'a', vector: new Float32Array([1, 0, 0, 0]) }); await db.insert({ id: 'b', vector: new Float32Array([0, 1, 0, 0]) }); const [best] = await db.search({ vector: new Float32Array([1, 0.1, 0, 0]), k: 1 }); // best.id === 'a' // GNN layer (from @ruvector/gnn): inputDim, hiddenDim, heads, dropout const layer = new RuvectorLayer(4, 8, 2, 0.1); const h = layer.forward( new Float32Array([1, 0, 0, 0]), // node embedding [new Float32Array([0, 1, 0, 0]), new Float32Array([0, 0, 1, 0])], // neighbor embeddings new Float32Array([0.5, 0.5]) // edge weights ); // Float32Array of length hiddenDim (8) // Soft top-k over candidate embeddings const { indices, weights } = differentiableSearch( new Float32Array([1, 0, 0, 0]), [new Float32Array([1, 0, 0, 0]), new Float32Array([0, 1, 0, 0])], 1, // k 1.0 // temperature ); // Access-frequency-aware embedding compression const tc = new TensorCompress(); const packed = tc.compress(new Float32Array([0.1, 0.2, 0.3, 0.4]), 0.5); // JSON string const restored = tc.decompress(packed); // Float32Array ``` ES modules work too: `import { VectorDB, RuvectorLayer } from '@ruvector/node'`. A default export bundles the same functions. Vectors must be `Float32Array`. See the [`@ruvector/core` README](https://www.npmjs.com/package/@ruvector/core) for `VectorDB` options (distance metrics, `hnswConfig`, `storagePath` defaults) and the [`@ruvector/gnn` README](https://www.npmjs.com/package/@ruvector/gnn) for the GNN API. ## Exports | Entry point | Exports | |-------------|---------| | `@ruvector/node` | `VectorDB`, `RuvectorLayer`, `TensorCompress`, `differentiableSearch`, `hierarchicalForward`, `getCompressionLevel`, `initGnn`; types `VectorEntry`, `SearchQuery`, `CoreSearchResult`, `GnnSearchResult`, `CompressionLevelConfig` | | `@ruvector/node/gnn` | `RuvectorLayer`, `TensorCompress`, `differentiableSearch`, `hierarchicalForward`, `getCompressionLevel`, `init`; types `SearchResult`, `CompressionLevelConfig` | ## Related packages - [`@ruvector/core`](https://www.npmjs.com/package/@ruvector/core) — the vector database re-exported here; install it alone if you don't need GNN. - [`@ruvector/gnn`](https://www.npmjs.com/package/@ruvector/gnn) — the GNN layers re-exported here. - [`@ruvector/graph-wasm`](https://www.npmjs.com/package/@ruvector/graph-wasm) — property graph + Cypher in WebAssembly, for the relationships your GNN consumes. - [`ruvector`](https://www.npmjs.com/package/ruvector) — umbrella package and CLI (`npx ruvector`). ## License MIT ## Works with / Used by Part of the [ruv.io](https://ruv.io) open-source stack for graph neural network, Node.js, vector search. - **Works with** (its runtime dependencies): [`@ruvector/core`](https://www.npmjs.com/package/@ruvector/core), [`@ruvector/gnn`](https://www.npmjs.com/package/@ruvector/gnn) - **Used by** (published packages that depend on it): none yet --- Built by [rUv](https://ruv.io) · [ruv.io](https://ruv.io) · [Source](https://github.com/ruvnet/ruvector)