This page runs the rule-based --offline mode on a snapshot of the dataset: no LLM, no network. It runs the exact same command the repo's README tells you to run yourself, so this page and your own terminal can never disagree. The full agent uses Sanity Context and an LLM for freer-form questions; see the repo.