The DuckDB Universe: every project and company built on DuckDB

Open original

This is a free, searchable directory of 575 open-source projects and companies built on DuckDB, the embedded analytics database. It's rebuilt from the awesome-duckdb GitHub list and updated as of October 2026.

  • The map covers 516 open-source projects and 59 companies that publicly say they build on DuckDB, each linked to its source.
  • The most-starred projects include PyGWalker (16K stars), Kepler.gl (12.1K), and Lance (7.2K), and the top UIs are Preswald and Online DuckDB Shell.
  • Snowflake's own Snowset data shows 96.2% of queries scanned under 10 GB, a size one DuckDB instance can handle, so many workloads could move off a warehouse.

The DuckDB Universe is a searchable directory of 575 projects and companies built on DuckDB, the open-source embedded analytics database under the MIT license. Of those, 516 are open-source projects pulled from the awesome-duckdb GitHub list with archived repos dropped, and 59 are companies or products that publicly say they build on or with DuckDB, each linked back to its source. The page is organized into categories such as extensions, client-server setups, DuckDB clients and UIs, backends, and integrations, and it is rebuilt from that GitHub list, so adding a project means submitting it there. The snapshot is dated 2026-10-26. The most-starred open-source projects are PyGWalker (16K stars), Kepler.gl (12.1K), Lance (7.2K), Evidence (7.1K), Ibis (6.7K), AliSQL (6K), smallpond (5K), and Apple's Embedding Atlas (5K). On the company side, the list includes Airbyte, AWS, Cloudflare, dbt, Dagster, Deepnote, DeepSeek, and dlt, among others. For UIs, Preswald leads at 4.3K stars, followed by Online DuckDB Shell and Codapi. On the extensions front, DuckDB has 29 core and 267 community extensions, with stochastic, anofox_forecast, gsheets, h3, and mssql the most downloaded community extensions last week according to duckdb.org. Installing one is a single statement: INSTALL name FROM community. The most interesting claim is about Snowflake. The article leans on Snowflake's own Snowset dataset of 69 million customer queries from 2018, where the median query scanned just 5 MB and 96.2% scanned under 10 GB. That's a workload a single machine running DuckDB can handle comfortably. The few very large queries still justify a warehouse, which is the gap Greybeam fills by routing each Snowflake query to DuckDB or Snowflake depending on size. The data is a few years old and comes from one vendor's workload, so treat the "replace Snowflake" framing as a hypothesis worth testing on your own query logs rather than a settled fact.

Ask about this article