Boost AI Vector Search with MyScaleDB: ClickHouse‑Powered SQL Database

Overview

MyScaleDB is a SQL‑compatible vector database built on ClickHouse, designed to let developers create scalable AI applications using familiar SQL syntax. It combines vector search and storage with relational capabilities, handling large volumes of structured and unstructured data while reducing engineering complexity.

Key Features

Full SQL Compatibility

Fast, powerful vector search, filter search, and SQL‑based vector joins.

Use standard SQL functions for vector operations—no new tools or frameworks required.

Production‑Ready for AI

Unified platform for managing text, vectors, JSON, geospatial, time‑series, and other data types.

Combines vectors with rich metadata to enable high‑precision, high‑efficiency filtered search, improving RAG accuracy.

Unmatched Performance & Scalability

Leverages ClickHouse’s OLAP architecture and advanced vector algorithms for lightning‑fast vector calculations.

Scales economically as data grows, outperforming custom‑API vector databases and offering lower cost than pgvector or Elasticsearch extensions.

Why Choose MyScaleDB

Complete SQL compatibility.

Unified management of structured and vector data.

Millisecond‑level search on billions of vectors.

High reliability and linear scalability.

Support for hybrid search and complex SQL‑vector queries.

Built on ClickHouse, MyScaleDB benefits from columnar storage, advanced compression, skip indexes, and SIMD processing, making filter‑then‑vector search highly accurate and performant.

Architecture

MyScaleDB serves as the data backbone for next‑generation large‑model + big‑data solutions, providing high‑throughput SQL and vector capabilities for data processing, knowledge retrieval, observability, analytics, and few‑shot learning, forming a closed AI‑data loop.

Quick Start

Run the official Docker image to start a MyScaleDB instance:

docker run --name mysaledb myscale/myscaledb:1.4

The container starts with a default user default and no password. Connect with the ClickHouse client:

docker exec -it mysaledb clickhouse-client

Tutorial

Create a Table with a Vector Column

-- Create a table with body_vector of length 384
CREATE TABLE default.wiki_abstract (
    `id` UInt64,
    `body` String,
    `title` String,
    `url` String,
    `body_vector` Array(Float32),
    CONSTRAINT check_length CHECK length(body_vector) = 384
) ENGINE = MergeTree ORDER BY id;

Insert Data

-- Insert data from Parquet files on S3
INSERT INTO default.wiki_abstract SELECT * FROM s3('https://myscale-datasets.s3.ap-southeast-1.amazonaws.com/wiki_abstract_with_vector.parquet','Parquet');

Create a Vector Index

-- Build a SCANN vector index with Cosine metric on body_vector
ALTER TABLE default.wiki_abstract ADD VECTOR INDEX vec_idx body_vector TYPE SCANN('metric_type=Cosine');

-- Check index build progress
SELECT * FROM system.vector_indices;
-- Wait until the status becomes 'Built'

Perform a Vector Search

-- Return the top‑5 nearest vectors
SELECT id, title,
       distance(body_vector, [-0.052, -0.0146, -0.0677, ...]) AS distance
FROM default.wiki_abstract
ORDER BY distance ASC
LIMIT 5;

Tech Stack

ClickHouse – open‑source OLAP database for large‑scale analytics.

Faiss – Meta’s library for efficient similarity search and dense vector clustering.

hnswlib – C++/Python library for fast approximate nearest‑neighbor search.

ScaNN – Google Research’s scalable nearest‑neighbors library.