#browser-ai

HN treated Mozilla's Prompt API objection as a warning about browser-AI lock-in, focusing on model-specific prompt tuning, vendor terms, and the return of compatibility hacks.

HN saw the appeal immediately: local prompts, no API keys, more privacy. The thread turned just as quickly to the friction points, especially the storage and hardware bill attached to browser-side AI.

LocalLLaMA reacted with genuine wonder because the demo is simple to grasp: a 1.7B Bonsai model, about 290MB, running in a browser through WebGPU. The same thread also did the useful reality check, asking about tokens per second, hallucinations, llama.cpp support, and whether 1-bit models are ready for anything beyond narrow tasks.