Lakehouse 2 min read

The SQL editor and notebooks, rebuilt: tabs, versions, and real .ipynb files

A multi-tab SQL editor where one tab runs while you write in another, with saved script versions and per-statement results — and notebooks that import and export genuine Jupyter files, so your existing work comes with you and your new work can leave.

Two of the surfaces you spend the most time in got rebuilt.

The SQL editor

Tabs that run independently. Start a long query in one tab and keep writing in another — each tab owns its own run state instead of the whole editor going busy. Cmd/Ctrl+T for a new tab, Cmd/Ctrl+W to close, Cmd/Ctrl+1–9 to jump.

Run all, with one result sub-tab per statement, so a script of six statements gives you six sets of results instead of only the last one. There's a real Cancel — including "stop all running queries" when you've started more than you meant to.

Scripts have versions. Saved server-side, with history you can look through and share, because "I had it working an hour ago" should be recoverable.

Once results are on screen, you can work with them without re-querying: filter, pick columns, or open Summary for per-column types, null counts, distinct values, and ranges. Export CSV or JSON, chart it, save the results to your lakehouse, or turn the statement straight into an automation.

Notebooks that speak Jupyter

The important thing about a notebook feature is whether your existing notebooks work in it and whether your new ones can leave.

Import accepts real nbformat 4.x files, mapping the kernel and keeping cell ids where the format carries them. Saved outputs are dropped — you get the code, and you re-run it against your own data. Export emits valid nbformat 4.5, which means nbconvert, nbdime, and GitHub's own renderer all read what comes out. Not a lookalike format; the actual one.

An older nbformat 3 file is rejected by name, with an explanation. The alternative — importing it as a silently empty notebook — is the kind of small dishonesty that costs someone an afternoon.

The full modal keyboard layer is there: Esc and Enter for command and edit modes, A/B to insert, J/K to move, Y/M to change cell type, C/X/V to copy, cut, paste, Z to undo, Shift+M to merge. Plus a cell command palette, cross-cell find and replace, a table of contents built from your markdown headings, and Run this section.

Cells run SQL, Python, PySpark, and Scala against your lakehouse.

When a cell breaks

An errored cell offers Explain this error and Fix this cell — and a fix arrives as a proposed change you can Accept, or Accept and run. It doesn't rewrite your cell behind your back.

Worth knowing

Notebooks carry a Beta badge, and they've earned the honesty that goes with it: Python cells can't be interrupted once running — SQL cells can — and all tabs in the editor share a single connection, so you can't put a Trino tab next to a Databricks tab.

Everything else is exactly what it looks like. Open the SQL editor, hit Cmd/Ctrl+T, and start.

NewerSnowflake key-pair authentication is here — before you need it → News & insights ← Agent skills, rewritten: see it, name it, own itOlder

See it on your own data — five minutes, $50 in credit, no card.