Vector Store

VectorStore in the spring-ai-vector-store module is the interface to manage and query documents in a vector database.

VectorStore

public interface VectorStore extends DocumentWriter {
  void add(List<Document> documents);

  void delete(List<String> idList);

  void delete(Filter.Expression filterExpression);

  default void delete(String filterExpression) {}

  List<Document> similaritySearch(SearchRequest request);
}

Create VectorStore

VectorStores are created using builders. Different VectorStore implementations have their own builder implementations. VectorStore.Builder defines common methods for VectorStore builders.

Property type	Method	Description
ObservationRegistry	observationRegistry	Registry for observation support
VectorStoreObservationConvention	customObservationConvention	Custom convention for creating observations
BatchingStrategy	batchingStrategy	Batching strategy to embed documents

Add Documents

Documents are added using the add method.

Add documents

vectorStore.add(docs);

Delete Documents

Documents are deleted using the delete methods. When deleting documents, we can provide a list of IDs of documents, or a filter expression to select documents. The filter expression can be specified using Filter.Expression or String.

Delete documents

vectorStore.delete("topic == 'AI'");

Search Documents

Documents are searched using the similaritySearch method. Search requests are represented using SearchRequest. SearchRequests are created using builders.

The table below shows methods of SearchRequest builder.

Property type	Method	Description	Default value
String	query	Text to use for embedding similarity comparison	""
int	topK	top k similar results to return	4
double	similarityThreshold	Similarity threshold score to filter the search response by	0.0
Filter.Expression	expression	Expression to filter documents	null
String	textExpression	Expression to filter documents, parsed by FilterExpressionTextParser	null

The code below shows how to search documents.

Search documents

vectorStore.similaritySearch(
  SearchRequest.builder()
    .query("Build agents")
    .topK(1)
    .build());