With a self-hosted LLM, that loop happens locally. The model is downloaded to your machine, loaded into memory, and runs ...