Offline LLM system runs entirely in browser

True air-gapped AI requires moving beyond local runtime to local-only reproduction and deployment, as most local LLM tools fail in high-security environments due to initial internet dependencies for model downloads. By bundling embeddings, models, and document chunks into a single exportable package, this system treats the entire AI stack as a static asset, using browser-native execution via Transformers.js and WebGPU to bypass complex server orchestration across different hardware.