Gemini API CLI does not plan tasks or actively invoke models; it merely encapsulates multimodal generation capabilities into a single command, with the coding Agent responsible for decisions and the CLI responsible for execution.