Skip to content

Search

Synopsis

kbolt search [OPTIONS] <QUERY>

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 --semantic are mutually exclusive.
  • The default search mode is hybrid retrieval without reranking unless --rerank is set.
  • --deep reranks by default; use --no-rerank if you want the broader retrieval path without the reranking pass.
  • --keyword and --semantic bypass the default hybrid path.
  • --keyword and --semantic always skip reranking.
  • normal CLI output groups chunk matches by document and may show +N more matching sections when multiple hits came from the same file
  • --format json returns the raw chunk-level SearchResponse from the engine without CLI grouping
  • --debug changes the text presentation layer so you can inspect raw chunk hits, pipeline stages, and per-signal scores

Examples

Default hybrid search:

kbolt search "rust error handling"

Default hybrid search with reranking:

kbolt search "rust error handling" --rerank

Keyword-only search:

kbolt search "E0502" --keyword

Semantic-only search:

kbolt search "partial index recovery" --semantic

Deep search:

kbolt search "index corruption recovery" --deep

Scoped search:

kbolt --space work search "deploy checklist" --collection runbooks