I was dickin' around with with Clawbot/Openclaw but ran into your dilemma. A month or so later I realize that the system I'd built for CAD, 3D print workflow, and fooling with virtual machines would be half decent with a couple of RTX pro 2000's thrown into the mix.
I run Hermes on a VM and it uses qwen3.6-35b-a3b on LM Studio on the host machine (sandboxing that lil' fucker.)
(cont)
