No model loadedConfigure the sidebar
Brimkern · Local WebGPU inference
A standalone, optimized build powered by hand-written WGSL compute shaders. Your models and computations run entirely locally, with no third-party server.
Downloaded once, kept on this device: next visits start in seconds. 100% local.
step 1
Pick a model
One click is enough — the weights stream in. Or drag and drop your own GGUF (Qwen, Gemma, Llama…).
step 2
Compute on the GPU
The JS parser extracts the tensors and our WGSL kernels run the forward pass live.
Try any model from Hugging Face
Single-file GGUF and .brik, straight from the Hub. Nothing to configure: the best quantization is picked, and the tokenizer follows the file. Nothing leaves your browser.
Examples:
Brimkern — open WebGPU engine, built by Romain Khanoyan. Local AI, WebGPU, on-device engines.
Brimkern runs in isolation and never sends your data anywhere.