from what I understand it just makes it easier to swap models, but it still uses llama.cpp under the hood.