Starring repositories is collecting evidence. Asking questions of that collection is how the evidence pays rent.
That is the job of Starcat's knowledge-base RAG: retrieve from your indexed Stars, then answer with citations you can open and verify.
Not another generic chat window
A general-purpose model already knows popular frameworks. It does not know:
- which of your Stars you marked as "using"
- the private note where you wrote "too heavy for our CLI"
- the README chunk you indexed last week for a niche library nobody blogs about
RAG closes that gap by grounding retrieval in the local knowledge base you built from Stars, notes, and related context.
Citations are the product
An answer without sources is a vibe. An answer with repository citations is a workflow: click through, skim the README, decide.
Starcat leans on explainable retrieval — what was searched, what was kept, what was cited — so you can distrust a bad hop without throwing away the whole feature.
Start with a real question
Good first prompts look like work, not demos:
- "Which starred repos help with local SQLite full-text search on macOS?"
- "What did I save related to OpenSSF / supply-chain scoring?"
- "Summarize options I starred for README rendering."
If the library is empty of structure, RAG will feel thin. If you have even a modest set of Stars with readable READMEs indexed, questions start to land.
The point is not to replace reading. It is to find where to read next.