iris-vector-search

name	iris-vector-search
description	Implement IRIS native vector similarity search. Use when adding semantic search, embedding storage, or nearest-neighbor queries to an IRIS application.
license	MIT
source	intersystems-community/iris-vector-rag
metadata	{"version":"1.0.0","author":"InterSystems Developer Community","compatibility":"iris, objectscript, python"}

Purpose

Use IRIS SQL VECTOR type and HNSW index for fast semantic similarity search — no external VectorDB.

When to Use

• Storing and searching embeddings in IRIS globals/tables
• Adding VECTOR_COSINE or VECTOR_DOT_PRODUCT search to existing ObjectScript classes
• Replacing an external VectorDB with native IRIS capability

Core SQL Patterns

Store embeddings

INSERT INTO MyEmbeddings (content, embedding)
VALUES (?, TO_VECTOR(?, FLOAT, 1536))

Search by cosine similarity

SELECT TOP 10 content,
    VECTOR_COSINE(embedding, TO_VECTOR(?, FLOAT, 1536)) AS score
FROM MyEmbeddings
ORDER BY score DESC

Create HNSW index

CREATE INDEX emb_hnsw ON MyEmbeddings (embedding)
WITH (TYPE='HNSW', M=16, EF_CONSTRUCTION=200, DISTANCE='COSINE')

ObjectScript Access

Set stmt = ##class(%SQL.Statement).%New()
Do stmt.%Prepare("SELECT TOP ? content, VECTOR_COSINE(embedding, TO_VECTOR(?, FLOAT, 1536)) score FROM MyEmbeddings ORDER BY score DESC")
Set rs = stmt.%Execute(topK, embeddingVector)
While rs.%Next() { Write rs.content, " (", rs.score, ")", ! }

Dimension Guide

Model	Dimensions
text-embedding-3-small	1536
text-embedding-3-large	3072
nomic-embed-text	768
fastembed BAAI/bge-small	384