So you can use llama.cpp and point a harness to it? What harness do you use?