so, how are people running these locally? is there a llama.cpp/ollama solution for systemone/jev style apis?