GlassLLM

A playful LLM visualizer. Watch a tiny language model generate tokens in your browser while attention, KV similarity, layer activity, and top-k predictions light up as it thinks.

Runs entirely client-side — no backend, no API key, no database.

Quick start

npm install
npm run dev

Open the Vite URL, type a prompt, and press run. The first run downloads the model into your browser cache.

Commands

npm run build   # type-check + bundle
npm test        # run tests
npm run lint    # type-check only