DuckDB

DuckDB is an analytical SQL database management system that can run in-process and be deployed across environments ranging from edge devices to servers with hundreds of cores. It uses a PostgreSQL-inspired SQL dialect and can read and write common data formats, including CSV, JSON, Parquet, Iceberg, Delta Lake, Avro, Arrow, Excel, Vortex, Lance and DuckLake, either from local storage or from object storage. Queries can be executed directly against files, including remote Parquet and CSV files accessed over HTTPS.

The project is positioned as part of a broader "Duck Stack" that includes DuckDB itself, Quack (a client-server protocol in which both client and server are full DuckDB instances, allowing remote database access alongside client-side computation), DuckLake (a lakehouse format built on SQL that stores data in object storage and uses a SQL database as catalog), and first-class Apache Iceberg support for reading and writing Iceberg tables. Functionality can be extended through extensions that add functions and support for additional formats.

DuckDB provides native client APIs for major programming languages and interfaces, including a command line interface, Python, Go, Java (JDBC), Node.js, C, C++, R, Rust and ODBC, with installation through package managers such as pip, npm, cargo, go get, downloadable artifacts, and a shell install script. It also integrates with data frame libraries such as pandas and R tooling including dplyr and duckplyr, and supports user-defined functions and bulk loading via an Appender API.

Business model

DuckDB and related projects are distributed under the MIT license as free and open-source software, governed by the independent DuckDB Foundation.

Latest developments

An August 2026 blog post by Mark Raasveldt and Hannes Mühleisen announced that DuckLabs would join AWS while the projects remain open source. Recent engineering blog posts covered DuckDB table functions in Java and recursive CTE execution performance.