Spaces:
Running
Running
| title: vROM.js Demo | |
| emoji: 🔍 | |
| colorFrom: indigo | |
| colorTo: purple | |
| sdk: static | |
| pinned: false | |
| license: mit | |
| short_description: Browser vector search with WASM HNSW + ONNX | |
| # 🔍 vROM.js Demo | |
| An interactive landing page and live demo for [vROM.js](https://www.npmjs.com/package/vrom.js) — a zero-boilerplate RAG SDK for browser AI agents. | |
| ## What This Demo Shows | |
| - **Live semantic search** over HuggingFace documentation using vROM.js | |
| - **vROM hot-swapping** between different knowledge bases | |
| - **Performance metrics** (embedding time, HNSW search time) | |
| - **Code examples** showing the 5-line API | |
| - **Architecture diagrams** explaining how it works | |
| ## How It Works | |
| Everything runs 100% in your browser: | |
| 1. The WASM engine (172 KB) handles HNSW vector search in < 1ms | |
| 2. A Web Worker runs ONNX embedding inference via transformers.js | |
| 3. Pre-computed vROM indexes are loaded from HF Hub CDN and cached in OPFS | |
| 4. No server, no API keys, no backend — just open the page and search | |
| ## Links | |
| - **[npm package](https://www.npmjs.com/package/vrom.js)** — `npm install vrom.js` | |
| - **[GitHub](https://github.com/philsupertramp/vecdb-wasm)** — Source code | |
| - **[Full Documentation](https://huggingface.co/spaces/philipp-zettl/vecdb-wasm)** — API reference, guides, architecture | |