What it gives you
- Vector and hybrid language filters match chunk language; lexical search accepts the language option but ignores it
- Automatic query expansion finds related terms you didn’t search for
- Default seed-scoped graph expansion adds filtered one-hop linked candidates after first fusion
- Reranking pushes the most relevant results to the top
- Search several collections at once (-c work -c ops); the collection filter applies before the result limit
- Date filters (--since, --until) for time-sensitive work
- Tag, category, author, and language filters stay active when graph retrieval is enabled
- Intent steering to bias results toward a topic
- Exclusion filters to remove noise
- Structured query modes for power users (term, intent, hyde)
- Trusted local project affinity adds one bounded, explainable workspace signal
- contentTypes searchBoost adds a bounded query-agnostic prior without bypassing filters
- Opt-in typed metadata from gno.metadata frontmatter: filter on strings, numbers, booleans, and arrays, with coverage warnings for invalid or pending fields
- Chunk size and overlap set per index, with applied-policy status, so two indexes over one corpus can compare chunking policies
- Every retrieval stage is configurable