I've started creating https://github.com/g023/localmodelrouter/ which offers Ollama like functionality but as a single .py file with minimal dependencies and more focus on letting llama.cpp handle the dirty work.