Definition

What it is

Embedded OLAP SQL engine powered by ClickHouse. In-process analytical database for Python, Go, Rust, Node.js. Queries Parquet, Arrow, CSV, JSON directly without external server.

Recent developments

Latest signals

ClickHouse 25.8 Vector Search GA. ClickHouse 25.8 brings vector search with binary quantization to general availability. Per clickhouse.com (2026-01-20).
chDB Kernel Upgrade to ClickHouse v25.8. chDB upgrades its embedded ClickHouse kernel to v25.8 achieving 61x performance improvement. Per clickhouse.com (2025-11-19).
chDB v4 shipped a zero-copy Pandas DataStore API — the headline release the current node is missing. chDB v4 (landed March 2026) introduces a pandas-compatible DataStore API spanning 630+ methods with true zero-copy integration against Pandas DataFrames, powered by ClickHouse v25.8.2.1. Per chDB docs and ClickHouse January 2026 newsletter.
Zero-copy delivers large, vendor-published but methodologically detailed speedups. ClickHouse's own benchmark reports 247x faster than Pandas on COUNT(*), ~24% faster than DuckDB when exporting 1M rows to Pandas (2.64s vs 3.47s), and 87x faster than chDB v1. Per chDB: a fast, reliable, and scalable in-process database and The Journey to Zero-Copy.
Narrower platform support than DuckDB. chDB v4 is Python-only and limited to macOS/Linux — no Windows, no other language bindings, unlike DuckDB's broader ecosystem reach. Per Embedded Databases in 2026: DuckDB, SQLite, Polars, and chDB.
Latest release v4.1.9 (June 9 2026) improves Parquet-over-S3 reading. Recent fixes include RTLD_DEEPBIND preload handling, SQL builder unification, memory optimization, and Parquet S3 reading — directly relevant to lakehouse-style Parquet-on-S3 workflows. Per chdb-io/chdb releases. Sources: clickhouse.com · clickhouse.com

Connections 2

Outbound 2

scoped_to1

Lakehouse

depends_on1

ClickHouse