Live · needs WebGPU (Chrome/Edge) and a ~1.5 GB model download
A Prior-Auth Review Agent, Entirely in Your Browser
A mini agent harness in JavaScript driving a fine-tuned Qwen3-0.6B over WebGPU — tool calls, criteria citations, approve/deny/needs-more-info decisions, all on your GPU. Nothing leaves your machine: the page downloads the weights (~1.5 GB) and every token is generated locally. Synthetic data only; 30% → 77.5% over the stock model.