We're actually exploring that with our app, if you've a GPU large enough you can run a llama2 based model locally at almost chatgpt speeds... If you're interested in trying drop me a message at rafael (at) curiosity.ai