Back to blog
Announcements05 Nov 263 mins

Polars Cloud becomes Polars Nanook

Polars Cloud becomes Polars Nanook as the platform matures beyond cloud managed compute to a full data platform for both cloud and on-premise (airgapped) environments.

Ritchie Vink, Chiel Peters

Share article

When we launched Polars Cloud last year, one often asked question was “Can it run in my own environment / infrastructure?”. Polars was already powering production workloads, but for large scale queries clients still fell back to PySpark or a home-grown partitioning setup. We built on-premises to support these questions and along the way the product grew beyond compute with query optimizations & profiling. “Cloud” became a misnomer and we renamed the product Polars Nanook. For all Polars Cloud users today, nothing changes and your existing setup will continue to work. See the section Migration on how to easily migrate to the new package name.

Migration

From now on you can install pip install polars-nanook instead of pip install polars_cloud and use import polars_nanook as pn. The cloud package will continue to work for the upcoming transition period of 6 months and will now give a FutureWarning indicating you to transition to the new package. Any existing code will continue to work during this period.

Today’s release

Besides the name change, today also marks a new release of Polars Nanook:

  • New website: With the new name comes a completely redesigned website that covers both open source Polars and Polars Nanook in one place, explaining our vision around the full platform. Go visit at pola.rs.
  • SQL support in Nanook: As of Polars 2.0 SQL is now supported as a first class citizen next to our existing DataFrame API. SQL coverage has drastically improved and that extends to Nanook as well. To run SQL simply add remote behind the SQLContext.
  • Distributed engine support: Polars has a very large API surface. This release marks a huge leap in lowering most of Polars’ API to distributed algorithms, including non-trivially distributable operations like group_by_dynamic, asof_joins, rolling_agg, etc. Furthermore, this release drastically increases the resilience to OOM by shipping with out-of-core support enabled.

What’s next?

As we mature Polars’ compute engine and distributed capabilities, we are extending Nanook to become a data platform that will be tuned for agentic workloads. We will focus on managing the data sources, the governance and the semantic layer. In the following month(s) we’ll launch:

  • Data storage: First class Iceberg catalog support inside the platform allowing for easier data discovery, faster metadata serving & better query plans.
  • Warm compute: Polars Nanook currently spins up EC2 machines in your environment. For batch jobs the ~45s boot times are fine, however for interactive / agentic workflows these can be perceived as slow. With this option you can get compute instances in a single second and get started quickly using hot compute on our end.
  • Agentic access: All of the product features behind a single MCP, allowing users to write queries, spin up compute and discover their data from one common interface.

Try it out today!

Your Polars code stays the same. Nanook adds the platform features your enterprise needs: the cluster, the profiling and the overview of what your teams are running, in our cloud or in yours, airgapped if needed.

Start at nanook.pola.rs or contact us for a demo.