Like Ollama, I can use a feature-rich CLI, plus Vulkan support in llama.cpp and it takes a lot less disk space, too.