Run AI models locally on your machine with node.js bindings for llama.cpp. Force a JSON schema on the model output on the generation level