Answers are built only from records a real search returned. The assistant is instructed to refuse rather than invent a title, author, or finding.
AI repository assistant
Ask your repository anything
A grounded assistant that searches real published records, cites its sources, and tells you plainly when it does not know — never a fabricated answer.
Running in search-only mode on this installation — results are real published records; connect an AI provider for generated, cited answers.
How it works
One assistant, built to search first and never guess
Grounded search
Every answer starts with a real search
Before generating a word, the assistant searches published repository records for your question. The answer it gives you is built only from what that search actually found — titles, authors, years and excerpts — never invented.
Follow-up questions
Ask a follow-up without repeating yourself
The last few turns of your conversation are kept for context, so "what about more recent ones?" or "just the open-access records" work the way a real conversation should — each answer is still re-grounded in a fresh search, not carried forward blindly.
Honest by default
No match found means no match found
When a search turns up nothing relevant, the assistant says so directly and suggests a better term — it does not pad the answer with plausible-sounding filler. This works identically whether or not an AI provider is configured; without one, you still get real, ranked search results and an honest summary.
Responsible by design, not by promise
Questions and answers are never written to the application log. Only anonymous timing and status metrics are recorded, for reliability — not the content.
Public requests are throttled per visitor. A single actor cannot exhaust the assistant or run up cost on every other visitor's behalf.
Only published, publicly accessible metadata is searched. Restricted, embargoed, and private records are excluded from every search, every time.
Bring this to your own repository
Every Inrep Scholar deployment includes this assistant, grounded in your institution's own published records.
Request institutional proposal