[simonw]

You can even run a full LLM in your browser these days - try https://webllm.mlc.ai/ in Chrome, it can load up a Llama-2-7b chat model (~4000MB, took my connection just under 3 minutes) and you can start chatting with it.