A horde that answers your spreadsheet
"Who were our best customers last quarter?"

tableski turns your workbook into SQL tables that an AI can query. Kowalski, the open-source agent runtime tableski was hatched in, now ships the other half: a Spreadsheet analyst that takes your questions in plain words and hands back a report workbook. Version 2.2 went out today.
What happens when you ask
You type your questions, one per line. Six small agents (kowalski calls them penguins) take it from there:
- Your questions are recorded.
- Profile the tables: tableski lists every table, its columns, types and statistics. No model involved.
- Write the SQL: the model turns each question into one query, using the exact names from the profile.
- Run the queries through tableski. No model involved.
- Build the workbook:
report.xlsx, one sheet per question, numbers as numbers. No model involved. - Write the answers: short answers that quote the results they are handed.
The model writes queries and sentences. It never adds up a column. On our sample workbook, "who spent the most in total?" came back as Ada, 150.5, with the SQL that produced it saved next to the answer. "How many people live in each city?" came back as not answerable: the data has no city column, which is the answer you want when the data really cannot say.
Try it
Kowalski runs on your own Mac or Linux machine, with a free local model (Ollama) or your own key for a hosted one:
curl -fsSL https://raw.githubusercontent.com/yarenty/kowalski/main/install.sh | bash
kowalski
The Setup screen asks three questions; the third is your tableski account, signed in with one click. Load a workbook at tableski.io/app, pick the Spreadsheet analyst, and ask.
The full story, with a short recording and two more hordes (a morning brief and an inbox for invoices), is on the kowalski blog: Ask your spreadsheets. Get a workbook back. No number made up.