Vector search with filters in Neo4j v2026.01 (Preview)

Neo4j v2026.01 introduces a preview feature for vector search with filters, allowing users to apply predicates within the vector index at query time, optimizing latency and relevance without excessive data fetching. This feature, available across Neo4j Enterprise, Community, and Aura editions, aims to refine search results by filtering based on user criteria, such as language or category, directly in the index. It offers three filtering methods: in-index filtering for simple properties, post-filtering using Cypher to refine or expand results, and pre-filtering to define a candidate subgraph for exact scoring. Additionally, native Cypher syntax for vector search simplifies query authoring by integrating advanced search capabilities directly into Cypher, eliminating the need for procedure calls. Performance testing has shown that vector search with in-index filtering maintains low latency and high recall accuracy, with pre-filtering delivering 100% recall accuracy for small candidate sets but potentially higher computational costs for larger ones. The feature is part of Neo4j's efforts to enhance graph and search capabilities for AI and GraphRAG workloads, with general availability targeted for the next version and feedback encouraged to refine the offering.