The issue is not about running a local ai it is about where do you store your memory and context.
Anyway when you run an ai the thinking happens in the servers Openai or Claude of Google etc...