Skip to Content
Relevan Beta is here 🎉
BlogAgents can now learn an index's schema on the fly

Agents can now learn an index’s schema on the fly

2026-08-06

Relevan’s whole pitch is search built for agents, not just people clicking through a UI. That falls apart quickly if an agent has to be hand-taught your index’s field names and query syntax before it can search anything useful.

So every index now serves its own skill doc:

GET /v1/{orgId}/indexes/{indexName}/skill

It returns Markdown, generated fresh from that index’s live mapping — not a static doc that drifts out of sync the first time someone changes a field. An agent can fetch it once, right before it starts searching, and know:

  • Auth — the x-api-key header it needs.
  • How to search — the POST .../search request shape, with a reminder that filters are exact-match only and there’s no sort override in v1.
  • The fields on this specific index — a table of every field and its capabilities (searchable, filterable, sortable, aggregatable, autocompletable, highlightable), pulled straight from the mapping.
  • A worked example — a real request built from one of the index’s actual filterable fields, not a placeholder.

If the index doesn’t have a mapping deployed yet, the doc says so plainly instead of pretending there’s something to query.

This sits alongside llms.txt, which points agents at the API as a whole — the OpenAPI spec, auth, and the shape of every route. The skill doc is the zoomed-in version: everything an agent needs to know about this one index, in one fetch, with nothing to memorize ahead of time.

See the interactive API reference  for the rest of the API.

Last updated on