Expand description
Query-side filtering: a typed predicate tree over a point’s JSON payload.
How a filter runs is chosen by the embeddable database’s planner: when the
filter is selective on secondary-indexed fields it pre-filters to an exact
candidate scan, and otherwise it post-filters the vector-search candidates
(see quiver-embed and docs/index/design.md). Either way the Filter
tree is the stable wire shape and is re-checked on every surviving candidate,
so results are exact.
Field references are dot-paths into the payload object ("user.age").
Re-exports§
pub use sparse::DEFAULT_RRF_K0;pub use sparse::SPARSE_KEY;pub use sparse::SparseVector;pub use sparse::rrf_fuse;pub use sparse_index::BM25_B;pub use sparse_index::BM25_K1;pub use sparse_index::SparseInvertedIndex;pub use tokenize::TEXT_KEY;pub use tokenize::query_term_ids;pub use tokenize::term_id;pub use tokenize::text_to_sparse;pub use tokenize::tokens;
Modules§
- sparse
- Sparse vectors and Reciprocal Rank Fusion for hybrid search (ADR-0043).
- sparse_
index - Derived in-memory inverted index for sparse vectors (ADR-0045).
- tokenize
- A small, dependency-free tokenizer for the BM25 / full-text path (ADR-0046).
Enums§
- Filter
- A predicate over a point’s JSON payload.