neo4j-vector-index-skill

To create a vector index in Neo4j, use the CREATE VECTOR INDEX command with correct embedding dimensionality and similarity function alignment, then poll the index status until it reports ONLINE before querying nearest neighbors.

The best way to ingest embeddings into Neo4j graph nodes is via batch ingestion using UNWIND to set vector properties, or by using in-Cypher embedding with ai.text.embed(), ensuring dimension-safety checks are applied.

To perform vector similarity search in Neo4j 2026.01 and later, query nearest neighbors using the SEARCH clause, optionally combining results with graph traversal logic for hybrid retrieval patterns.

You can query vector indexes in older Neo4j versions by using the db.index.vector.queryNodes() procedure as a fallback to retrieve semantic nearest-neighbor results when the SEARCH clause is unavailable.

Neo4j vector search supports hybrid retrieval patterns by pairing vector similarity search results with graph traversal logic, enabling structural nearest-neighbor retrieval alongside semantic ranking.

Configuring Neo4j vector indexes requires setting the correct embedding dimensions, selecting an appropriate similarity function, and applying HNSW quantization options to optimize nearest-neighbor search performance.