barq-vweb

barq-vweb is a high-performance, browser-native vector database designed for modern web applications. It brings elite vector search capabilities—HNSW indexing, hybrid BM25+vector retrieval, and SIMD acceleration—directly to the client-side, with zero server-side dependencies for search.

---

Architecture

graph TD
    API[JS/TS API Layer] --> Search[Search Engine]
    API --> Embed[Embedding Engine]
    
    subgraph "barq-vweb"
        Embed --> MiniLM[MiniLM-L6-v2 Worker]
        Search --> Hybrid[Hybrid BM25 + RRF]
        Search --> HNSW[HNSW Index]
        HNSW --> PQ[Product Quantization]
    end
    
    subgraph "Acceleration & Storage"
        Bridge[compute/bridge.rs] --> BW[barq-wasm SIMD/WebGPU]
        Store[storage/IDB + OPFS] --> Disk[(IndexedDB / OPFS)]
    end
    
    HNSW --> Bridge
    HNSW --> Store

Loading

Key Features

• Self-Contained Embeddings: Integrated all-MiniLM-L6-v2 via Transformers.js—runs in a Web Worker to keep your UI fluid.
• Elite Indexing: Professional HNSW implementation ported from barq-db for sub-linear query times.
• Hybrid Retrieval: Out-of-the-box support for combining keyword-based BM25 search with semantic vector search using Reciprocal Rank Fusion (RRF).
• Hardware Acceleration: Seamless integration with barq-wasm for 16-wide SIMD unrolled kernels and WebGPU/WebNN dispatches.
• Persistent Storage: Native OPFS (Origin Private File System) and IndexedDB backends for reliable client-resident data.

---

Quickstart

Installation

npm install barq-vweb

Usage

import { BarqVWeb } from "barq-vweb";

// 1. Initialize DB with persistence
const db = new BarqVWeb("my-collection");
await db.init(); // Loads WASM, barq-wasm SIMD, and Embedding Model

// 2. Insert text (Automatic embedding + HNSW indexing)
await db.insertTexts([
  "Rust is a systems programming language focused on safety.",
  "WebAssembly enables near-native speed in the browser."
]);

// 3. Perform Hybrid Search
const results = await db.search("fast systems programming", { limit: 5 });
console.log(results);

---

Performance Benchmarks

Operation	barq-vweb (SIMD)	Voy	Voy (JS)	EdgeVec
Insert 1k vectors	~9ms	~28ms	~45ms	~38ms
kNN-search (10k)	~1.2ms	~9ms	~18ms	~14ms
Embed single text	~8ms	N/A	N/A	N/A
Batch embed 100 texts	~180ms	N/A	N/A	N/A

Tests conducted on a Mac M2 Pro. Vector search dimension = 384.

---

Browser Support

Technology	Chrome	Firefox	Safari	Edge
WebAssembly	✅ 113+	✅ 115+	✅ 17+	✅ 113+
WebGPU	✅ 113+	🔬 Flag	✅ 17+	✅ 113+
OPFS	✅ 102+	✅ 111+	✅ 15.2+	✅ 102+
Web Worker	✅ All	✅ All	✅ All	✅ All

---

Building from Source

Ensure you have Rust and wasm-pack installed.

# Clone the repository
git clone https://github.com/YASSERRMD/barq-vweb.git
cd barq-vweb

# Build the WASM package and JS bindings
bash build.sh

# Run example development server
cd examples
npm install
npm run dev

---

License

Licensed under MIT.