Data Analyst

Data Analyst answers questions about a CSV or Excel file you attach — with charts and a short note on how it worked the answer out — and cleans or extends a workbook by handing back an edited copy with a change log. Your original file is never changed.

At a glance

How it works

1. It sees the shape of your data first

   When you attach a spreadsheet, the agent is given its outline — every sheet, each column with its type, the row count and the first rows — so the analysis is right the first time.

2. The analysis runs in the sandbox

   Your file is copied into an isolated sandbox with no network access, and the agent works on it there with pandas. Nothing it writes can leave except the results it hands back.

3. Charts from the computed result

   A summary table the analysis produces becomes a chart directly — the numbers are never retyped — and you can explore it like any chart in chat.

4. A Method note, and the script

   Every answer ends with which sheet and rows were used, what was excluded and why, and how it was calculated. The script that produced it comes back as analysis.py, so "show me how" is a download.

Cleaning or changing a workbook

Ask it to "clean up the dates", "add a margin column" or "remove the duplicate rows" and you get <your file> (edited).xlsx back. It is edited in place on a copy, so formatting and formulas it did not touch survive, and it carries a Change log sheet listing every change — sheet, cell or column, before, after and why.

How the router finds it

Attach a spreadsheet and ask one question about it, and the request reaches the agent tagged data before any model is asked. A request that also asks for something else — "analyse this and book me a flight" — is planned across agents instead, so nothing you asked for is dropped. Without an attachment, the router decides as usual.