The problem is the "for free" part. Unless you have the hardware for it at home (and it has to be very good hardware) the locally run LLMs don't have enough capacity for the context windows that these agents need to be most effective so you have to hit one of the bigger providers APIs which gets exp...