Private, on-device grammar and spell checking powered by local small language models with WebGPU acceleration.
