Dark Inference · WebGPUA model on your own GPU. Nothing leaves the tab.
Open weights compiled for WebGPU load into this page once, then every token is computed by your graphics card. There is no API behind the box below, and the network witness counts every request the page makes after the model is ready.
Your GPUAsking the browser…
Model
Network witnessArms itself once the weights are loaded. From then on, every request the page makes is counted here.
Ask anything. Tell no one.Load a model, then write. The answer is computed by your own graphics card. There is no server behind this box.