Search¶
Synopsis¶
Arguments¶
<QUERY>: the search query
Options¶
--collection <COLLECTIONS>: restrict search to specific collections as a comma-separated list--limit <LIMIT>: maximum number of results to return--min-score <MIN_SCORE>: filter out results below a score threshold--deep: query expansion plus multi-variant retrieval--keyword: keyword-only search--semantic: dense-vector-only search--no-rerank: skip cross-encoder reranking--rerank: enable reranking on the default auto mode--debug: show pipeline stages and per-signal scores
Behavior notes¶
--deep,--keyword, and--semanticare mutually exclusive.- The default search mode is hybrid retrieval without reranking unless
--rerankis set. --deepreranks by default; use--no-rerankif you want the broader retrieval path without the reranking pass.--keywordand--semanticbypass the default hybrid path.--keywordand--semanticalways skip reranking.- normal CLI output groups chunk matches by document and may show
+N more matching sectionswhen multiple hits came from the same file --format jsonreturns the raw chunk-levelSearchResponsefrom the engine without CLI grouping--debugchanges the text presentation layer so you can inspect raw chunk hits, pipeline stages, and per-signal scores
Examples¶
Default hybrid search:
Default hybrid search with reranking:
Keyword-only search:
Semantic-only search:
Deep search:
Scoped search: