visual RAG text → image, live CLIP

Type a query in plain words — CLIP ViT-B/32 embeds it client-side and ranks a corpus of real document screenshots by visual + semantic meaning. Text and pixels share one embedding space, all in your browser.