It's very straightforward. Think of it as creating a common 'bus' to connect LLMs to daemon/service running on your system. `pull` models, `run` models, interact with them via CLI, GUI.