Private by design
Ask your data.
| name | class | sex | age | fare | survived | embarked |
|---|---|---|---|---|---|---|
| Allen, Miss. Elisabeth | 1 | female | 29 | 211.34 | 1 | Southampton |
| Braund, Mr. Owen | 3 | male | 22 | 7.25 | 0 | Southampton |
| Cumings, Mrs. John | 1 | female | 38 | 71.28 | 1 | Cherbourg |
| Heikkinen, Miss. Laina | 3 | female | 26 | 7.92 | 1 | Southampton |
| Futrelle, Mrs. Jacques | 1 | female | 35 | 53.10 | 1 | Southampton |
| Allen, Mr. William | 3 | male | 35 | 8.05 | 0 | Southampton |
| McCarthy, Mr. Timothy | 1 | male | 54 | 51.86 | 0 | Southampton |
| Palsson, Master. Gosta | 3 | male | 2 | 21.08 | 0 | Southampton |
| Johnson, Mrs. Oscar | 3 | female | 27 | 11.13 | 1 | Southampton |
| Nasser, Mrs. Nicholas | 2 | female | 14 | 30.07 | 1 | Cherbourg |
| Sandstrom, Miss. Marguerite | 3 | female | 4 | 16.70 | 1 | Southampton |
| Bonnell, Miss. Elizabeth | 1 | female | 58 | 26.55 | 1 | Southampton |
| Saundercock, Mr. William | 3 | male | 20 | 8.05 | 0 | Southampton |
| Andersson, Mr. Anders | 3 | male | 39 | 31.28 | 0 | Southampton |
| Hewlett, Mrs. Mary | 2 | female | 55 | 16.00 | 1 | Southampton |
| Rice, Master. Eugene | 3 | male | 2 | 29.13 | 0 | Queenstown |
| Williams, Mr. Charles | 2 | male | 23 | 13.00 | 0 | Southampton |
| Vestrom, Miss. Hulda | 3 | female | 14 | 7.85 | 0 | Southampton |
| Fortune, Mr. Charles | 1 | male | 19 | 263.00 | 0 | Southampton |
| Masselmani, Mrs. Fatima | 3 | female | 22 | 7.23 | 1 | Cherbourg |
| Fynney, Mr. Joseph | 2 | male | 35 | 26.00 | 0 | Southampton |
| Beesley, Mr. Lawrence | 2 | male | 34 | 13.00 | 1 | Southampton |
| McGowan, Miss. Anna | 3 | female | 15 | 8.03 | 1 | Queenstown |
| Sloper, Mr. William | 1 | male | 28 | 35.50 | 1 | Southampton |
It never leaves your browser.
Curator runs the whole analysis right on your device — your rows and values never touch a server. It writes the Python, runs it on your data locally, and shows you every line behind the answer. Private, and provable.
Your data never leaves your device · No file upload · See every line of code
Verifiable, not just plausible
A number you can change and re-check.
Open the code behind any answer, edit a line — drop a filter, exclude a segment — and hit Run. The chart and the number update from your edit, not a black box. That’s the difference between a guess and an answer.
How it works
Three steps, nothing hidden.
Bring your data
Drop in a CSV, Excel, JSON, or Google Drive file. Every sheet becomes a table; load several and Curator finds the keys to join them.
Ask in plain English
q = df[df.quarter.isin(["Q3", "Q4"])]g = q.groupby(["region", "quarter"]).revenue.sum()g.unstack().pct_change(axis=1)
Your question becomes real Python, run in a sandbox inside your own browser tab. Your full file never leaves the page.
Get the answer + its code
Every result carries its proof one click away — the exact code, the rows it touched, the full trace. Edit a line and re-run anytime.
Built to be shared
A whole notebook, curated into one report.
When the analysis is done, turn the whole notebook into a clean, shareable report. One click for an AI summary of the findings — keep the code in for a technical audience or leave it out for a plain-English read — then export to PDF or share a link.
Do it once, then never again
Save an analysis. Run it on next week’s file.
Turn a finished notebook into a repeatable workflow. Drop in the new export and every step re-runs on the fresh data — same questions, same checks, new numbers. No rebuilding from scratch.
Your data, handled honestly
It runs in your browser — and we’re straight about the rest.
The honest version, not the marketing one: your file stays on your machine. The model sees your question and column names to write the code — never your rows, and your values are masked even when it phrases the answer.
Runs on your machine
Your file is loaded and analyzed in a sandboxed worker inside your own browser tab. The full dataset is never uploaded to, or stored on, a Curator server — there’s no ingest queue because the data never leaves.
- ✓ your question
- ✓ column names & types
- ✗ your rows
- ✗ your data values
Honest about what’s sent
To turn your question into Python, Curator sends the question and your column names and types — never your rows. When it phrases your answer in words, the values are masked before sending and restored on your machine, so your numbers never leave. See what’s sent →
The compounding part
Verified building blocks, not one-off guesses.
A chatbot regenerates the analysis on every question. Curator reuses code that’s already been checked — and run thousands of times.
correlation_pairIt reuses verified code
Ask the same question twice and you get the same trusted answer. Curator reuses building blocks that have already been checked and run — instead of re-writing the analysis, and re-rolling the risk, every time.
revenue_by_regioncertifiedcohort_retentionreviewedpct_change_by_groupreviewedseasonality_indexcommunity
It gets sharper over time
Every analysis can promote a block — community → reviewed → certified. The trusted library grows as it’s used, so reliability climbs over time instead of drifting.
Questions
The things people ask first.
How is this different from asking ChatGPT?
Do I need to know how to code?
What can I upload, and how big can it be?
Is my data private?
What does it cost?
Pricing
Simple, and free to start.
- 30 questions a month
- CSV, Excel, JSON & Google Sheets
- Join across multiple tables
- Every chart with its code, trace & data
- Share a clean report (export to PDF)
- Unlimited questions
- Everything in Free
- Curated reports — AI summary & custom pick-and-refine
- Cancel anytime
Ask your data something hard.
Put a real question to your spreadsheet — every answer comes with its receipts.
Open the workbenchNo sign-up · Free to start