Add filter-based deletion to VectorStore implementations, Chroma, Elastic, PGVector

spring-ai-core/src/main/java/org/springframework/ai/vectorstore/VectorStore.java

@@ -1,5 +1,5 @@
 /*
- * Copyright 2023-2024 the original author or authors.
+ * Copyright 2023-2025 the original author or authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -24,9 +24,11 @@
 import org.springframework.ai.document.Document;
 import org.springframework.ai.document.DocumentWriter;
 import org.springframework.ai.embedding.BatchingStrategy;
+import org.springframework.ai.vectorstore.filter.Filter;
 import org.springframework.ai.vectorstore.observation.DefaultVectorStoreObservationConvention;
 import org.springframework.ai.vectorstore.observation.VectorStoreObservationConvention;
 import org.springframework.lang.Nullable;
+import org.springframework.util.Assert;
 
 /**
  * The {@code VectorStore} interface defines the operations for managing and querying
@@ -62,6 +64,28 @@ default void accept(List<Document> documents) {
 	@Nullable
 	Optional<Boolean> delete(List<String> idList);
 
+	/**
+	 * Deletes documents from the vector store based on filter criteria.
+	 * @param filterExpression Filter expression to identify documents to delete
+	 * @throws IllegalStateException if the underlying delete causes an exception
+	 */
+	void delete(Filter.Expression filterExpression);
+
+	/**
+	 * Deletes documents from the vector store using a string filter expression. Converts
+	 * the string filter to an Expression object and delegates to
+	 * {@link #delete(Filter.Expression)}.
+	 * @param filterExpression String representation of the filter criteria
+	 * @throws IllegalArgumentException if the filter expression is null
+	 * @throws IllegalStateException if the underlying delete causes an exception
+	 */
+	default void delete(String filterExpression) {
+		SearchRequest searchRequest = SearchRequest.builder().filterExpression(filterExpression).build();
+		Filter.Expression textExpression = searchRequest.getFilterExpression();
+		Assert.notNull(textExpression, "Filter expression must not be null");
+		this.delete(textExpression);
+	}
+
 	/**
 	 * Retrieves documents by query embedding similarity and metadata filters to retrieve
 	 * exactly the number of nearest-neighbor results that match the request criteria.

spring-ai-core/src/main/java/org/springframework/ai/vectorstore/observation/AbstractObservationVectorStore.java

@@ -1,5 +1,5 @@
 /*
- * Copyright 2023-2024 the original author or authors.
+ * Copyright 2023-2025 the original author or authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -18,6 +18,7 @@
 
 import java.util.List;
 import java.util.Optional;
+import java.util.stream.Collectors;
 
 import io.micrometer.observation.ObservationRegistry;
 
@@ -27,6 +28,7 @@
 import org.springframework.ai.vectorstore.AbstractVectorStoreBuilder;
 import org.springframework.ai.vectorstore.SearchRequest;
 import org.springframework.ai.vectorstore.VectorStore;
+import org.springframework.ai.vectorstore.filter.Filter;
 import org.springframework.lang.Nullable;
 
 /**
@@ -99,6 +101,18 @@ public Optional<Boolean> delete(List<String> deleteDocIds) {
 			.observe(() -> this.doDelete(deleteDocIds));
 	}
 
+	@Override
+	public void delete(Filter.Expression filterExpression) {
+		VectorStoreObservationContext observationContext = this
+			.createObservationContextBuilder(VectorStoreObservationContext.Operation.DELETE.value())
+			.build();
+
+		VectorStoreObservationDocumentation.AI_VECTOR_STORE
+			.observation(this.customObservationConvention, DEFAULT_OBSERVATION_CONVENTION, () -> observationContext,
+					this.observationRegistry)
+			.observe(() -> this.doDelete(filterExpression));
+	}
+
 	@Override
 	@Nullable
 	public List<Document> similaritySearch(SearchRequest request) {
@@ -131,6 +145,19 @@ public List<Document> similaritySearch(SearchRequest request) {
 	 */
 	public abstract Optional<Boolean> doDelete(List<String> idList);
 
+	/**
+	 * Template method for concrete implementations to provide filter-based deletion
+	 * logic.
+	 * @param filterExpression Filter expression to identify documents to delete
+	 */
+	@Nullable
+	protected void doDelete(Filter.Expression filterExpression) {
+		// this is temporary until we implement this method in all concrete vector stores,
+		// at which point
+		// this method will become an abstract method.
+		throw new UnsupportedOperationException();
+	}
+
 	/**
 	 * Perform the actual similarity search operation.
 	 * @param request the search request

vector-stores/spring-ai-chroma-store/src/main/java/org/springframework/ai/chroma/vectorstore/ChromaApi.java

@@ -338,7 +338,7 @@ public AddEmbeddingsRequest(String id, float[] embedding, Map<String, Object> me
 	 */
 	@JsonInclude(JsonInclude.Include.NON_NULL)
 	public record DeleteEmbeddingsRequest(// @formatter:off
-		@JsonProperty("ids") List<String> ids,
+		@Nullable @JsonProperty("ids") List<String> ids,
 		@Nullable @JsonProperty("where") Map<String, Object> where) { // @formatter:on
 
 		public DeleteEmbeddingsRequest(List<String> ids) {

vector-stores/spring-ai-chroma-store/src/main/java/org/springframework/ai/chroma/vectorstore/ChromaVectorStore.java

@@ -25,6 +25,8 @@
 import com.fasterxml.jackson.core.JsonProcessingException;
 import com.fasterxml.jackson.databind.ObjectMapper;
 import com.fasterxml.jackson.databind.json.JsonMapper;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
 
 import org.springframework.ai.chroma.vectorstore.ChromaApi.AddEmbeddingsRequest;
 import org.springframework.ai.chroma.vectorstore.ChromaApi.DeleteEmbeddingsRequest;
@@ -40,6 +42,7 @@
 import org.springframework.ai.vectorstore.AbstractVectorStoreBuilder;
 import org.springframework.ai.vectorstore.SearchRequest;
 import org.springframework.ai.vectorstore.VectorStore;
+import org.springframework.ai.vectorstore.filter.Filter;
 import org.springframework.ai.vectorstore.filter.FilterExpressionConverter;
 import org.springframework.ai.vectorstore.observation.AbstractObservationVectorStore;
 import org.springframework.ai.vectorstore.observation.VectorStoreObservationContext;
@@ -81,6 +84,8 @@ public class ChromaVectorStore extends AbstractObservationVectorStore implements
 
 	private boolean initialized = false;
 
+	private static final Logger logger = LoggerFactory.getLogger(ChromaVectorStore.class);
+
 	/**
 	 * @param builder {@link VectorStore.Builder} for chroma vector store
 	 */
@@ -162,7 +167,29 @@ public Optional<Boolean> doDelete(@NonNull List<String> idList) {
 	}
 
 	@Override
-	public @NonNull List<Document> doSimilaritySearch(@NonNull SearchRequest request) {
+	protected void doDelete(Filter.Expression expression) {
+		Assert.notNull(expression, "Filter expression must not be null");
+
+		try {
+			ChromaFilterExpressionConverter converter = new ChromaFilterExpressionConverter();
+			String whereClauseStr = converter.convertExpression(expression);
+
+			Map<String, Object> whereClause = this.chromaApi.where(whereClauseStr);
+
+			logger.debug("Deleting with where clause: {}", whereClause);
+
+			DeleteEmbeddingsRequest deleteRequest = new DeleteEmbeddingsRequest(null, whereClause);
+			this.chromaApi.deleteEmbeddings(this.collectionId, deleteRequest);
+		}
+		catch (Exception e) {
+			logger.error("Failed to delete documents by filter: {}", e.getMessage(), e);
+			throw new IllegalStateException("Failed to delete documents by filter", e);
+		}
+	}
+
+	@Override
+	@NonNull
+	public List<Document> doSimilaritySearch(@NonNull SearchRequest request) {
 
 		String query = request.getQuery();
 		Assert.notNull(query, "Query string must not be null");

vector-stores/spring-ai-chroma-store/src/test/java/org/springframework/ai/chroma/vectorstore/ChromaVectorStoreIT.java

@@ -1,5 +1,5 @@
 /*
- * Copyright 2023-2024 the original author or authors.
+ * Copyright 2023-2025 the original author or authors.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -36,6 +36,7 @@
 import org.springframework.ai.openai.api.OpenAiApi;
 import org.springframework.ai.vectorstore.SearchRequest;
 import org.springframework.ai.vectorstore.VectorStore;
+import org.springframework.ai.vectorstore.filter.Filter;
 import org.springframework.boot.SpringBootConfiguration;
 import org.springframework.boot.test.context.runner.ApplicationContextRunner;
 import org.springframework.context.annotation.Bean;
@@ -169,6 +170,69 @@ public void addAndSearchWithFilters() {
 		});
 	}
 
+	@Test
+	public void deleteWithFilterExpression() {
+		this.contextRunner.run(context -> {
+			VectorStore vectorStore = context.getBean(VectorStore.class);
+
+			// Create test documents with different metadata
+			var bgDocument = new Document("The World is Big and Salvation Lurks Around the Corner",
+					Map.of("country", "Bulgaria"));
+			var nlDocument = new Document("The World is Big and Salvation Lurks Around the Corner",
+					Map.of("country", "Netherlands"));
+
+			// Add documents to the store
+			vectorStore.add(List.of(bgDocument, nlDocument));
+
+			// Verify initial state
+			var request = SearchRequest.builder().query("The World").topK(5).build();
+			List<Document> results = vectorStore.similaritySearch(request);
+			assertThat(results).hasSize(2);
+
+			// Delete document with country = Bulgaria
+			Filter.Expression filterExpression = new Filter.Expression(Filter.ExpressionType.EQ,
+					new Filter.Key("country"), new Filter.Value("Bulgaria"));
+
+			vectorStore.delete(filterExpression);
+
+			// Verify Bulgaria document was deleted
+			results = vectorStore
+				.similaritySearch(SearchRequest.from(request).filterExpression("country == 'Bulgaria'").build());
+			assertThat(results).isEmpty();
+
+			// Verify Netherlands document still exists
+			results = vectorStore
+				.similaritySearch(SearchRequest.from(request).filterExpression("country == 'Netherlands'").build());
+			assertThat(results).hasSize(1);
+			assertThat(results.get(0).getMetadata().get("country")).isEqualTo("Netherlands");
+
+			// Clean up
+			vectorStore.delete(List.of(nlDocument.getId()));
+		});
+	}
+
+	@Test
+	public void deleteWithStringFilterExpression() {
+		this.contextRunner.run(context -> {
+			VectorStore vectorStore = context.getBean(VectorStore.class);
+
+			var bgDocument = new Document("The World is Big", Map.of("country", "Bulgaria"));
+			var nlDocument = new Document("The World is Big", Map.of("country", "Netherlands"));
+			vectorStore.add(List.of(bgDocument, nlDocument));
+
+			var request = SearchRequest.builder().query("World").topK(5).build();
+			assertThat(vectorStore.similaritySearch(request)).hasSize(2);
+
+			vectorStore.delete("country == 'Bulgaria'");
+
+			var results = vectorStore.similaritySearch(request);
+			assertThat(results).hasSize(1);
+			assertThat(results.get(0).getMetadata().get("country")).isEqualTo("Netherlands");
+
+			vectorStore.delete(List.of(nlDocument.getId()));
+		});
+	}
+
 	@Test
 	public void documentUpdateTest() {
 

vector-stores/spring-ai-elasticsearch-store/src/main/java/org/springframework/ai/vectorstore/elasticsearch/ElasticsearchVectorStore.java

@@ -26,6 +26,7 @@
 import co.elastic.clients.elasticsearch.ElasticsearchClient;
 import co.elastic.clients.elasticsearch.core.BulkRequest;
 import co.elastic.clients.elasticsearch.core.BulkResponse;
+import co.elastic.clients.elasticsearch.core.DeleteByQueryResponse;
 import co.elastic.clients.elasticsearch.core.SearchResponse;
 import co.elastic.clients.elasticsearch.core.bulk.BulkResponseItem;
 import co.elastic.clients.elasticsearch.core.search.Hit;
@@ -221,6 +222,23 @@ public Optional<Boolean> doDelete(List<String> idList) {
 		return Optional.of(bulkRequest(bulkRequestBuilder.build()).errors());
 	}
 
+	@Override
+	public void doDelete(Filter.Expression filterExpression) {
+		// For the index to be present, either it must be pre-created or set the
+		// initializeSchema to true.
+		if (!indexExists()) {
+			throw new IllegalArgumentException("Index not found");
+		}
+
+		try {
+			this.elasticsearchClient.deleteByQuery(d -> d.index(this.options.getIndexName())
+				.query(q -> q.queryString(qs -> qs.query(getElasticsearchQueryString(filterExpression)))));
+		}
+		catch (Exception e) {
+			throw new IllegalStateException("Failed to delete documents by filter", e);
+		}
+	}
+
 	private BulkResponse bulkRequest(BulkRequest bulkRequest) {
 		try {
 			return this.elasticsearchClient.bulk(bulkRequest);