NotesQL
Join the waitlist

Pre-launch · early access

The SQL notebook where Claude is your data analyst.

Write documents that mix notes, SQL, results and charts. Ask a question in plain words and Claude drafts the query from your schema and your team's own definitions. You read it. You click Run.

Join the waitlist See how it works
  • Postgres
  • MySQL
  • CSV & Parquet via DuckDB
Growth Weekly revenue review analytics · read-only

Weekly revenue review

Finance asked whether September was soft. Checking net revenue by plan.

Which plans grew net revenue in September?

Draft · not run Ran after review
metric: net_revenue join: invoices → accounts 
SELECT a.plan,
  SUM(CASE WHEN i.month = '2026-09'
      THEN i.net_amount END) AS sep,
  SUM(CASE WHEN i.month = '2026-08'
      THEN i.net_amount END) AS aug
FROM invoices i
JOIN accounts a ON a.id = i.account_id
GROUP BY a.plan ORDER BY sep DESC;
Run Explain Edit
3 rows212 msrun by you
plansepaugchange
Business$48.2k$41.9k+15%
Team$31.6k$33.0k−4%
Starter$12.4k$11.8k+5%

Business carried September, up 15% on August. Team was the only plan that fell.

Cites net_revenue, defined by Finance

Illustration with sample data.

How it works

Ask, review, run. In that order, every time.

Claude does the drafting. A person makes the call. The query that touches your database is always one somebody has read.

  1. 01

    Ask

    Type a question into the doc, the way you would ask a colleague. Claude reads your schema and your documented metrics, joins and example queries, then drafts the SQL.

    How many trials converted last week?
  2. 02

    Review

    The draft lands as an ordinary SQL block, marked as not run. Read it, ask Claude to explain it in plain words, or edit it yourself.

    Draft · not runSELECT COUNT(*) FROM trials …
  3. 03

    Run

    Nothing executes until you click Run, over a read-only connection. Results, a suggested chart and a short write-up land in the same doc.

    Runread-only · 1 row

What Claude does in the notebook

  • Drafts SQL from a plain-language question, using your schema and definitions.
  • Explains a query in plain words, line by line if you want.
  • Repairs a failing query from the database error, as a new draft for you to review.
  • Suggests a chart that fits the shape of the result.
  • Writes the narrative around the numbers, so the doc reads like a report.

Governed context

Answers come from your definitions, not a guess.

Ask three people what "active customer" means and you get three queries. NotesQL keeps one shared layer of meaning next to your data: the metrics your team has agreed on, how tables join, the words people actually use for them, and queries someone has checked.

Claude drafts from that layer first, and every answer says which definitions it used. When a definition changes, the next draft follows it.

  • Metrics with an owner and a written definition.
  • Joins declared once, so nobody fans out a table by accident.
  • Synonyms, so "sales" and "bookings" land on the right metric.
  • Verified queries that Claude can learn the house style from.
metricnet_revenueverified
Definition
Invoiced amount minus refunds and credits, by paid date.
Owner
Finance
Also called
revenuesalesnet sales
SUM(i.amount - i.refunds - i.credits)
joininvoices → accounts

invoices.account_id = accounts.id · many to one

“Net revenue was $92.2k in September.”

Cites net_revenue · invoices → accounts

Safety by design

Claude drafts. You decide what runs.

The guardrails are part of how NotesQL is built, not a setting you have to remember.

Nothing runs without a click

Claude never executes SQL by itself. Every generated query is shown for review and runs only when a person clicks Run.

Read-only connections

Postgres and MySQL go through a read-only query gateway. A notebook can read your data; it can't change it.

Answers cite their sources

Each answer names the metric definitions and joins it relied on, so you can check the reasoning, not just the number.

Built to share safely

The aim for sharing: readers see saved results, never a live query against your database, with personal data masked by default. It's being built alongside the Claude analyst.

Who it's for

For people who have questions about their data.

Analysts

Skip the boilerplate. Let Claude draft the routine queries and first-pass write-ups while you check the logic and keep the definitions honest.

Product and ops teams

Ask in your own words and get a query you can read, built on the metrics your data people already agreed on.

Startups without a data team

Point it at Postgres, MySQL or a CSV export and get real answers, with read-only access and a review step before anything runs.

Where it stands

Pre-launch, and plain about it.

Built

  • The notebook: text, SQL blocks, results and charts in one document
  • Read-only query gateway for Postgres and MySQL
  • Querying uploaded CSV and Parquet files with DuckDB
  • Sharing notebooks

In development

  • Claude as your analyst: drafting, explaining, repairing, charting and writing
  • The semantic layer: metrics, joins, synonyms and verified queries
  • Safe sharing: saved results only, personal data masked by default

Get early access

Early access opens in small groups once the Claude analyst is ready. Send a line about your data and the questions you keep asking it, and we'll write when there's a seat.

Join the waitlist

Built on Claude by Anthropic.