Vision Lab — VR

Point at a glyph in your headset, pull the trigger, and a vision-language model running on-device describes what's in front of you. Compatible candidates orbit an anchored star in your view, sized and tilted by their real classifier-supplied orbital parameters. Frames never leave the browser.

…WebXR (immersive-vr)
…WebGPU
…Origin Private File System (cache)

SMOLVLM-256M ~250 MB · one-time download · cached by service worker

Press Begin to download the model. Subsequent visits load in seconds.

Find candidates routes via the Meridian MCP — no PAT, no setup. Llama-3.3-70B authors candidates and the orbital classifier ranks them server-side.

No bundler — modules load from esm.sh on first hit and are then cached by the browser. Open in the Quest browser, complete the model download, then press the VR button.