Oh, that's just the infra for the infra. Then use something like graphllm from matteo, and of course llama.cpp from greg, tailor you model selection to your hardware.