Let Claude write the next notebook cell.

Before you start

You'll need a Flow Books notebook open against a connection, and an API key added in Settings if you want Claude to write cells rather than typing them by hand.

Step by step

1. Open Flow Books and create or open a notebook against your connection.
2. Add a SQL cell and run it to pull a result.
3. Add a Python cell below it and open Ask (or ⌘L).
4. Describe what you want the cell to do, mentioning the prior cell’s result if it should use it.
5. Review the code Claude writes, then run the cell.

A worked example: SQL into a chart

Pull a result with a SQL cell against a Redshift connection:

SELECT date_trunc('week', event_date) AS week,
       count(*) AS signups
FROM public.signups
GROUP BY 1
ORDER BY 1;

Then, on a Python cell, ask Claude to chart the signups-by-week result as a line plot. It writes a pandas/matplotlib cell using that result, so you're not exporting to a CSV and reimporting it into a separate notebook just to plot one number.

Check it worked

Run the cell. If the chart reflects the real numbers from the SQL cell above it rather than placeholder data, it picked up the right result. If it can't find what you're referring to, mention the prior cell more specifically, by what it queried, not just "the last result."

Troubleshooting

What this is good for

Quick joins between a SQL result and something else, a CSV, a second query, a hand-typed list, without leaving the notebook. It's not meant to replace a full data science environment; it's meant to remove the export-and-reimport step between a query and a chart or a quick calculation.

Where this fits next to the SQL editor's Ask panel

Flow Books and the SQL editor's Ask panel use the same underlying model and key, they're two surfaces for the same feature. Use the editor for a single query and its answer; use Flow Books when the next step is Python, a chart, a join with a file, or a few queries chained together in one document.

A second worked example: joining a query with a CSV

Say you have a SQL cell pulling customer IDs and lifetime value from Postgres, and a separate CSV of account managers by customer ID that isn't in the database at all. Load the CSV as a connection, then ask a Python cell to join the two on customer ID and show the top account manager by total lifetime value. Claude writes the pandas merge using both sources, something you'd otherwise do by exporting one side to a spreadsheet and doing the lookup by hand.

Keeping notebooks reproducible

Since each cell's code is visible and editable, a notebook Claude helped write isn't a black box, anyone opening it later can read exactly what ran, adjust a filter, and re-run it. That matters more the longer a notebook sticks around as a working reference rather than a one-off scratchpad.