Vector Searches

Vector searches use the HNSW index for fast retrieval, providing approximate results with a typical recall rate of over 95%.

---

Syntax

Following is an example using the Euclidean distance:

SELECT ID, <vector_column_name> <-> '[0.1,0.2,0.3...]' as distance FROM <table_name> ORDER BY <order_column_name> <-> '[0.1,0.2,0.3...]' LIMIT <top_k>;

---

Usage notes

• If no index is created, approximate searches will default to exact searches, which are slower but ensure 100% accuracy.

• To utilize vector index acceleration, the ORDER BY clause must include operators like <->, <#>, or <=>, paired with the appropriate vector index and distance metric. Without this pairing, acceleration cannot be applied. For more details on supported operators and usage, refer to Vector Indexing.

---

Examples

This example uses the knowledge base from Vector Indexing to retrieve source documents through vector-based text searches.

Following is the query text:

SELECT id, chunk, intime, url, embedding <-> '[10,2.0,..., 1536.0]' as distance FROM docs
ORDER BY
    embedding <-> '[10,2.0,..., 1536.0]'
LIMIT 100;