Guides

Notebooks

Interactive, git-backed notebooks that run on compute in your own cloud. Explore data, prototype models, and promote the same notebook straight into a scheduled job.

Git-backed

Notebooks live in a git repo cloned onto the agent. The agent keeps a single shared clone per repo; browser edits are working-tree changes until you commit. Switching branches checks out on the agent.

Because the working tree is shared per repo on the agent, GitOps and job runs read from origin, commit and push to make changes durable and reproducible.

Run cells

Attach a notebook to a running compute instance and execute cells against a live kernel. Queries to the lakehouse route through your SQL warehouse exactly as they do from the CLI.

Reuse as a task

A notebook can be referenced directly by a job task, so the code you prototyped becomes a scheduled, retryable step with no rewrite. See Jobs & pipelines.

polnor jobs create nightly-eda \
  --task eda --notebook notebooks/eda.ipynb --compute training-a100

Track experiments

From inside a notebook, log metrics and params with the Python SDK, it's MLflow-compatible. See Models & serving.

from polnor import mlflow
mlflow.log_metric("auc", 0.93)

Where a notebook is used

The console shows back-references, every job task and run that uses a given notebook, so you always know the blast radius before you change it.