I’ve installed koboldcpp on a thinkpad x1 with 32gb RAM and a i7-1355U, no GPU. Sure, it’s only just around 1 token/s but for a chat it is still usable (about 15 s per reply). The setup was easier than expected.