Crate lexa_core 

Structs

Chunk

Document

EmbeddingConfig

IndexStats

LexaDb

PreprocessOutput

Output of Preprocessor::preprocess.

SearchHit

SearchOptions

TierBreakdown

Transaction

Represents a transaction on a database connection.

Enums

EmbeddingBackend

LexaError

SearchTier

Retrieval tier — selects the work the search engine does for a query.

Constants

EMBEDDING_DIMS

Native embedding dimension. Tied to the model below; both must change together (and the existing vectors_bin schema must be re-indexed).

Traits

Preprocessor

Per-file callback signature used by index_path_with_preprocessor.

Functions

default_db_path

matryoshka_truncate

Truncate a Matryoshka-trained embedding to a smaller prefix and re-normalize. Nomic v1.5 is MRL-trained at canonical dims 64, 128, 256, 512, 768, so any prefix is a valid embedding in the same vector space. fastembed already returns L2-normalized embeddings; we re-normalize after truncation so cosine scores stay in [-1, 1].

open

Type Aliases

Result