I’ve set it up on my home computer for testing with code generation. My main tasks are machine learning projects, simple web apps, etc. I’ve connected it to the continue.dev extension in vs code and tried some of the smaller code models. I’m going to do some testing to find what completion models work the best, then fiddle with chat models. Agent mode was bad with the default small models, but then again I’ve never been impressed with agent mode even with GitHub copilot using 4o.