VEvectorize
Knowledge Base

Hybrid search over knowledge pages (BM25 + vector)

Changed on

Doc-level hybrid search across a bank's knowledge pages: a full-text (BM25) match and a vector-similarity match, Reciprocal-Rank-Fusion fused. No reranker — tuned for latency.

get/v1/default/banks/{bank_id}/knowledge-base/search

Request

Path parameters

bank_idstring required

Query parameters

qstring required

Search query

limitinteger

Maximum results to return

tagsstring[] nullable

Only return pages carrying these tags (matched per tags_match, like recall).

tags_match'any' | 'all' | 'any_strict' | 'all_strict' | 'exact'

How tags match a page's tags: any, all, any_strict, all_strict, exact. 'any'/'all' also return untagged pages; the _strict modes and 'exact' do not.

tag_groupsstring nullable

JSON-encoded compound tag filter, same shape as recall's tag_groups, e.g. [{"or":[{"tags":["user:kate"],"match":"all_strict"},{"tags":["team"]}]}]. Top-level groups are AND-ed, and AND-ed with tags.

Headers

authorizationstring nullable

Response

Successful Response

totalinteger required

Changes