Your research retrieval
Connect your own retrieval system. By default GRITT sends only the question you submit here and your configured corpus label. You can include earlier messages only when your administrator allows it and you select a nonzero number below. Uploads are not sent to that API.
GRITT’s usual answer model, learning behaviour and subscription still apply. This connection adds evidence; it does not replace all GRITT services or guarantee that data stays with your institution.
Loading your permissions…
Permissions and approval
Connection
API contract for your developer
One POST request; no redirects or retries. A public HTTPS endpoint on port 443 with public IPv4 DNS is required. Return HTTP 200 and application/json within the configured timeout. Echo request_id exactly. Return no more than top_k passages, with at most max_chars text characters in total. Do not return instructions or a generated answer.
Request:
{"version":1,"request_id":"UUID","query":"Question","corpus":"course-a","top_k":8,"max_chars":8000}
Response:
{"version":1,"request_id":"same UUID","matches":[{"id":"passage-1","title":"Source title","text":"Source passage","url":"https://university.edu/source"}]}
url is optional. An empty matches array is valid. GRITT never follows returned links. The API key and corpus must restrict your endpoint to the intended corpus: GRITT cannot verify the permissions inside your retrieval server.
Connection test
No GRITT retrieval surcharge for this test. Your own provider may charge you.
Chat using your retrieval
Each submitted question is sent to your approved endpoint. Regular GRITT usage charges remain applicable. No automatic retry if retrieval fails.
Retrieval charges
Recent connection status
Last ten calls, without question or passage contents. Refresh to update.