I travel a lot and got tired of ChatGPT being unusable on flights (no Wi‑Fi), so I built a browser-based AI chat that runs locally. It downloads a small open model once (cached in IndexedDB), then runs inference on-device via WebGPU in a Web Worker, so it works offline and nothing leaves your machine. Trade-offs: smaller models (not GPT‑4), first load is a big download, and older hardware/mobile can struggle. However it works, and it goes into the direction of open-source, AI, and smaller, smarter models. Curious for feedback on the UX for model downloads, and whether people think browser-local AI is a viable direction.