CREATE VECTOR INDEX¶

Synopsis¶

CREATE VECTOR INDEX index_name
ON source_table ( column_name [, ...] )
[ WITH ( property_name = expression [, ...] ) ]
[ UPDATING FOR predicate ]

Description¶

Create a vector search index on the specified columns of a source table.

The column_name list specifies the columns to index. Typically this includes an identifier column and an embedding (vector) column.

The optional WITH clause sets index properties such as:

• index_type: The type of vector index (for example, 'ivf_rabitq4')

• distance_metric: The distance metric to use (for example, 'cosine')

• index_options: Additional index configuration (for example, 'nlist=100000,nb=8')

• partitioned_by: Partition columns (for example, ARRAY['ds'])

The optional UPDATING FOR clause filters the source data using a predicate, similar to a WHERE clause. This is typically used to specify which partitions to build the index for.

Examples¶

Create a vector index on an embeddings table for specific date partitions:

CREATE VECTOR INDEX my_index
ON my_table(content_id, embedding)
WITH (
    index_type = 'ivf_rabitq4',
    distance_metric = 'cosine',
    index_options = 'nlist=100000,nb=8',
    partitioned_by = ARRAY['ds']
)
UPDATING FOR ds BETWEEN '2024-01-01' AND '2024-01-03'

See Also¶

CREATE TABLE, CREATE TABLE AS