Gives LLMs precise answers from pre-researched code across repositories and languages - instead of re-scanning files and rebuilding context every session