Find the model. Keep the question.
Describe what you need. The ranking happens in your browser against a catalog you already hold, so the query is never sent anywhere — not in a URL, not to an embedding API, not as ciphertext.
- Runs
- On device
- Network calls
- 0
- Wallet
- Not required
What stays here
The query text, the ranking, and the order of results. They are computed in this tab from a catalog the page already downloaded, so there is no search request to intercept, log or subpoena.
What the server still sees
That your browser fetched the catalog, and — if you open a result — which resource page you asked for next. Opening a model is a normal page request and we do not pretend otherwise.
What this is not
Ranking runs on a lexical model today, so matches are approximate and based on catalog descriptions rather than measured quality. Prompts you later send to a provider are readable by that provider.