core/vector_store/index library
Dart port of a subset of FAISS' vector-index toolkit.
Mirrors the FAISS Python/C++ API: an Index holds a collection of
fixed-dimensionality float32 vectors and supports add, optional
train, and k-NN search. All indexes are CPU-only for now and
operate on Float32Lists for cache-friendly numerics.
Classes
- Index
- Base class for all FAISS-style indexes.
- RangeSearchResult
- Result of a radius / range search — variable-length hits per query.
- SearchResult
- Result of a k-NN Index.search call.
- TopK
-
Bounded-size max-heap of
(score, id)pairs, used by every index to accumulate the top-kneighbours in a single pass.
Functions
-
dot(
Float32List a, Float32List b) → double -
Dot product between two
d-length float vectors. -
l2sq(
Float32List a, Float32List b) → double -
Squared L2 distance between two
d-length float vectors. -
toFloat32Rows(
List< List< rows, int d) → List<double> >Float32List> -
Convenience: convert a row-major
List<List<double>>to a list ofFloat32Listrows (validating dimensionality).