Vector Search

GhostSQL provides first-class support for vector embeddings and similarity search.

Vector Type

You can define columns with the VECTOR(n) type, where n is the number of dimensions.

CREATE TABLE embeddings (
    id INT,
    vec VECTOR(3)
);

Similarity Operators

GhostSQL supports standard PostgreSQL similarity operators:

• <->: L2 (Euclidean) distance.
• <=>: Cosine distance.

Example Query

-- Find the closest row using L2 distance
SELECT * FROM embeddings 
ORDER BY vec <-> '[0.1, 0.2, 0.3]' 
LIMIT 1;

Vector Functions

For compatibility, GhostSQL also supports functional syntax:

• L2_DISTANCE(v1, v2)
• COSINE_DISTANCE(v1, v2)

SELECT id FROM embeddings 
ORDER BY COSINE_DISTANCE(vec, '[0.9, 0.1, 0.0]') 
LIMIT 5;

Indexing (HNSW)

For large datasets, you can create an HNSW index to speed up nearest neighbor searches:

CREATE INDEX my_idx ON embeddings USING HNSW (vec) WITH (m=16, ef_construction=200);