Browse documentation

Guides

Search your sources

Find original passages with keyword, vector, or hybrid retrieval.

Choose a retrieval mode

Keyword search matches text. Vector search finds passages by embedding similarity. Hybrid search combines keyword and vector retrieval under the configured backend.

The Search playground lets you inspect passages before asking the answer model to generate a response.

  • Keyword: useful for names, terms, and exact wording.
  • Vector: useful for related meaning and alternative phrasing.
  • Hybrid: combines the enabled keyword and vector signals.

Make a search request

POST /api/search accepts query, optional mode, and optional limit. The following body asks for up to five hybrid results.

JSON
{
  "query": "support opening hours",
  "mode": "hybrid",
  "limit": 5
}

Interpret a result

Results include documentId, versionId, chunkId, text, and a retrieval score. Source offsets identify the original passage.

A nearest match is not necessarily a relevant answer. Score meaning depends on the retrieval backend and mode; a high-looking number is not an answer-confidence percentage.

Current sources, current access

Retrieval applies current access policy and authoritative publication checks. A staged document or an inaccessible source does not become valid evidence because it exists in a derived index.

If an expected source is missing, inspect publication and job status before changing the question.