Struct SearchRequest 

pub struct SearchRequest<'e> { /* private fields */ }

Implementations

impl<'e> SearchRequest<'e>

pub fn image_scan(self, scan: ImageScan) -> Self

Opt into experimental Photo Anchor candidate scanning for standalone images. Exact f32 reranking preserves returned score semantics. Requires SQLite; a bounded shortlist does not guarantee recall on arbitrary libraries.

pub fn kind_counts(self, enabled: bool) -> Self

Include counts of returned hits by locator kind; no second search is run.

pub fn roots(self, roots: impl IntoIterator<Item = ResourceId>) -> Self

Add provider-specific roots without filesystem path conversion.

pub fn query(self, query: impl Into<String>) -> Self

Search for another phrasing too. Each segment is ranked by its best score over all queries, so rephrasing widens recall without diluting scores.

pub fn paths<I, P>(self, paths: I) -> Self

where I: IntoIterator<Item = P>, P: AsRef<Path>,

Add filesystem-only paths, canonicalized locally at execution even with a custom provider. Use roots() for generic provider identities.

pub fn limit(self, n: usize) -> Self

pub fn lexical_weight(self, w: f32) -> Self

Override how much keyword rank counts against semantic rank in hybrid mode (semantic is 1). By default it depends on the query, see [keyword_weight]. Must be finite and non-negative; zero disables the keyword signal in hybrid mode. Valid overrides are ignored in other modes.

pub fn mode(self, mode: SearchMode) -> Self

pub fn min_score(self, s: f32) -> Self

Hide semantic matches below this cosine similarity. Keyword and file name matches are kept: their relevance is not a cosine.

pub fn include(self, glob: impl Into<String>) -> Self

pub fn exclude(self, glob: impl Into<String>) -> Self

pub fn discovery(self, opts: DiscoveryOptions) -> Self

pub fn previews(self, chars: Option<usize>) -> Self

Maximum preview length in characters, or None to skip reading previews.

pub fn progress(self, f: ProgressFn<'e>) -> Self

pub fn sync(self, sync: bool) -> Self

With false, search only what is already cached and embed nothing. For callers where a separate indexer (see watch) keeps the cache fresh, so a search never waits for a large first index. Changed files may be stale.

pub fn lexical_fallback(self, enabled: bool) -> Self

Allow a cached search to prepare an ephemeral keyword index when any supported resource is new or changed. Never waits for embedding or writes to the semantic cache. Providers opting into cached listing retain their indexed-only scope; their background indexer discovers new resources. Without this opt-in, sync(false) stays cache-only.

pub fn execute(self) -> Result<SearchResults>