The model runs here.

The hero above is a simulation — this is not. SmolLM2-135M, the smallest model we ship, streamed by @xybrid/web on your GPU or CPU. The prompt never leaves this tab.

smollm2-135m-instruct @xybrid/web · preview WebGPU when available · wasm fallback

136 MB one-time download, cached by your browser. Replies capped at 256 tokens — a demo safeguard.

Example output — run it to see your own

"The AI's voice is like a gentle breeze,
A soothing melody that calms the mind,
It whispers secrets to those who listen,
A reminder that the world is a vast and wondrous place."

demo.ts
import { XybridLlm } from '@xybrid/web';

const llm = await XybridLlm.load('…/model_metadata.json', {
  wasmPath: '/demo/llm-runtime',
  accelerator: 'auto',
});

const stream = llm.generateStream(
  'Write a haiku about AI that works offline.',
  { maxOutputTokens: 256 },
);
for await (const delta of stream) render(delta);

The exact calls this page makes — it rewrites as you edit the prompt. bindings/web ↗