Now my gaming PC has exposed an API for lm studio (qwen3.5-35b-a3b) and also a boxlite sandbox service. I suddenly thought, isn't this exactly what I previously mentioned about having the inference and computing environment (sandbox) in one place? The ideal state for AIDC is to access all LLMs in an isolated computing environment without having to configure a bunch of messy settings. For example, being able to directly access cf's Worker AI in cf's sandbox, but currently their support is just too slow, and they don't even have the qwen 3.5 model.