Nl Query

Runs a natural-language search.

The interpreted query and resolved structures are returned alongside the results so the user can see -- and trust or correct -- how their question was understood. The model's translation and name lookups are cached (best-effort, in Valkey), so a repeated question skips the model call; the database query always re-runs so results stay fresh. A Valkey outage falls back to live calls rather than failing the request.

With dry_run=true the question is translated and resolved but the database search is not executed -- useful for inspecting exactly what query would run.