AI in the browser

About the AI in the browser thread

On-device AI is the technical bet the whole portfolio rests on, and this cluster is where the bet gets examined honestly. Running a model in a browser tab means the user's data never moves, which resolves the privacy problem structurally instead of by promise, but it also means running inside a memory budget, on whatever GPU is present, over a model you had to download first.

The articles cover both sides: how WebGPU and WebAssembly make local inference viable, what genuinely runs locally versus what still needs a server, what free cloud AI tools actually cost you in data, and whether any of it works offline. The recurring conclusion is that "private by design" is an architecture claim you can verify by watching the network tab, not a policy claim you have to trust.