Create vector index

Vector indexes allow Neo4j to run similarity queries across the whole database. The example below shows how to create a vector index for movie embeddings, under the name of moviePlots.

Provide a dimension when creating a vector index as shown in the examples below, as it ensures that only vectors of that size are indexed and makes dimension mismatches fail explicitly at query time.

CREATE VECTOR INDEX moviePlots
FOR (m:Movie)
ON m.embedding
OPTIONS {indexConfig: {
    `vector.dimensions`: 384,  (1)
    `vector.similarity_function`: 'cosine'  (2)
}}

1	For embeddings generated with the SentenceTransformers model `all-MiniLM-L6-v2`. If you use a different model, update this value to match its embedding dimension.
2	The cosine similarity function is the most common choice. For more details and other options, see Vector indexes → Cosine and Euclidean similarity functions.

CREATE VECTOR INDEX moviePlots
FOR (m:Movie)
ON m.embedding
OPTIONS {indexConfig: {
    `vector.dimensions`: 1536,  (1)
    `vector.similarity_function`: 'cosine'  (2)
}}

1	For embeddings generated with the OpenAI model `text-embedding-ada-002`. If you use a different model, update this value to match its embedding dimension.
2	The cosine similarity function is the most common choice. For more details and other options, see Vector indexes → Cosine and Euclidean similarity functions.

For more information on vector indexes, see Cypher → Indexes → Vector indexes.