mongodb

type DocumentCollection interface {
	BulkWrite(
		ctx context.Context,
		models []mongo.WriteModel,
		opts ...options.Lister[options.BulkWriteOptions],
	) (*mongo.BulkWriteResult, error)
	Aggregate(
		ctx context.Context,
		pipeline any,
		opts ...options.Lister[options.AggregateOptions],
	) (*mongo.Cursor, error)
	DeleteMany(
		ctx context.Context,
		filter any,
		opts ...options.Lister[options.DeleteManyOptions],
	) (*mongo.DeleteResult, error)
	SearchIndexes() mongo.SearchIndexView
}

type Similarity string

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

type StoreConfig struct {
	// Collection is the MongoDB collection that holds the documents.
	// Required.
	Collection DocumentCollection

	// VectorIndexName is the Atlas Vector Search index name. It must
	// match an existing index (or one created by InitializeSchema).
	// Optional: defaults to [DefaultVectorIndexName].
	VectorIndexName string

	// EmbeddingPath is the field that holds the document embedding.
	// Optional: defaults to [DefaultEmbeddingPath] ("embedding").
	EmbeddingPath string

	// ContentField is the field that stores the original text.
	// Optional: defaults to [DefaultContentField].
	ContentField string

	// MetadataField is the sub-document field that holds metadata.
	// Optional: defaults to [DefaultMetadataField]. Metadata is always isolated
	// in this sub-document so user keys cannot collide with storage fields.
	MetadataField string

	// MetadataFieldsToFilter pre-declares the metadata keys that
	// should be indexed as filter fields in the Atlas search index.
	// Filtering on a metadata field requires the field to be listed
	// here when InitializeSchema is true.
	MetadataFieldsToFilter []string

	// EmbeddingModel produces vectors for the documents. Required.
	EmbeddingModel embedding.Model

	// DocumentBatcher batches documents before upsert. Required.
	DocumentBatcher vectorstore.Batcher

	// Dimensions is the embedding width written into a new search-index
	// definition. When zero and InitializeSchema is true, the store probes
	// EmbeddingModel.
	Dimensions int

	// Similarity selects the vector similarity function. Optional:
	// defaults to [SimilarityCosine].
	Similarity Similarity

	// NumCandidates controls the recall/perf tradeoff of the Atlas
	// $vectorSearch stage. It is a floor: a search never considers fewer
	// candidates than the results it must return. Optional: defaults to
	// [DefaultNumCandidates] (200), and must not exceed [MaxNumCandidates].
	NumCandidates int

	// InitializeSchema, when true, creates the Atlas vector-search
	// index if it doesn't already exist. Requires a connected Atlas
	// cluster.
	InitializeSchema bool
}