I run ollama as my model for continue.dev in VS Code. It works fine, but it often will suggest one block of code, then write a completely different block into my files. I have no idea if the difference comes from ollama or continue, but... I cannot trust that it will do what it said it would. So there is no flippin' way I'd trust an AI to run commands on my system, regardless of what it told me it intended to run.