Engineering demo
Run
A WebGPU readiness check, in your browser. No upload, no account, no log.
When the cognitive cores ship, this page becomes the in-browser playground. For now it reports your WebGPU adapter — the substrate the cores will run on.
webgpu Detecting…
adapter —
vendor —
limits —
benchmark idle
elapsed —
per dispatch —
throughput —
Measured on your hardware. Browser, driver, thermal state, and concurrent GPU load all matter.
fit INT8 fit estimate
max buffer —
at INT8 (1 B/param) —
at INT4 (0.5 B/param) —
Hardware bound only. Does not include kernel scratch, KV cache, or activation memory.
Cognitive cores arrive on the roadmap. The same readiness check will become a real local inference session — same page, same adapter, no API key.