sqlitevec

Documentation

Overview

Package sqlitevec stores chunk embedding vectors in the same SQLite database as the document metadata, using the sqlite-vec extension's vec0 virtual table. Search is exact (brute-force) K-nearest-neighbour over unit-normalized vectors, so L2 distance ranks the same as cosine similarity.

Index

• func OpenVectorStorage(ctx context.Context, path string, dimensions int) (storage.VectorStorage, error)
• type Store
  • func Open(ctx context.Context, path string, dimensions int) (*Store, error)
  • func (s *Store) Close() error
  • func (s *Store) Delete(ctx context.Context, chunkIDs []int64) error
  • func (s *Store) EnsureSchema(ctx context.Context) error
  • func (s *Store) Replace(ctx context.Context, embeddings []storage.ChunkEmbedding) error
  • func (s *Store) Search(ctx context.Context, query []float32, limit int) ([]storage.VectorHit, error)

Functions

func OpenVectorStorage

func OpenVectorStorage(ctx context.Context, path string, dimensions int) (storage.VectorStorage, error)

OpenVectorStorage opens the vector store; Open already prepares the schema. It returns the interface so a failure yields a nil storage.VectorStorage, not a typed nil.

Types

type Store

type Store struct {
	// contains filtered or unexported fields
}

func Open

func Open(ctx context.Context, path string, dimensions int) (*Store, error)

Open connects to the SQLite database at path and ensures the vec0 vector table exists. The vectors live in the same file as the documents/chunks tables.

func (*Store) Close

func (s *Store) Close() error

func (*Store) Delete

func (s *Store) Delete(ctx context.Context, chunkIDs []int64) error

func (*Store) EnsureSchema

func (s *Store) EnsureSchema(ctx context.Context) error

EnsureSchema creates the vec0 virtual table if it does not exist. The vector column is a fixed-length float32 list sized to the configured embedding dimensions.

func (*Store) Replace

func (s *Store) Replace(ctx context.Context, embeddings []storage.ChunkEmbedding) error

Replace upserts the given chunk vectors. Existing rows for the same chunk ids are removed first, then the normalized vectors are inserted, all in one transaction.

func (*Store) Search

func (s *Store) Search(ctx context.Context, query []float32, limit int) ([]storage.VectorHit, error)

Search returns the limit nearest chunk vectors to the query, closest first. The query is normalized so that L2 distance ranks the same as cosine similarity.