Notes from the rookery
- Latest 路 Kowalski 2.3
Ask more about your spreadsheet
Kowalski 2.3 uploads your workbook to tableski from the horde page, lets you ask a follow-up on the same tables, and reads sheets without a header row. The numbers still come from SQL.
- Kowalski 2.2
A horde that answers your spreadsheet
Kowalski 2.2 ships a Spreadsheet analyst: six penguins, four of them without a model, turn plain questions about your tableski tables into a report workbook. The numbers come from SQL, not from the AI.
- Demo
Five minutes from a spreadsheet to a conversation
The whole demo, step by step: a two-sheet workbook goes in, kowalski or Claude asks it questions in plain words, SQL does the work. Every step is a command you can copy for your own file.
- Speed
Fast enough for a conversation
We timed the hosted service on a 100,000-row spreadsheet and a million-row CSV, then timed the usual Python way of answering the same question. One number is 190x the other.
- Safety
An AI with SQL and good manners
Letting a language model run SQL over your data is a great idea right up to the first DROP TABLE. How tableski decides what a query may do before it runs.
- Under the hood
What happens to your workbook
Sheets become tables, headers get detected, dates stop being numbers, and the ugly files we collected on purpose still have to load.
