Ask your spreadsheet a question.

Upload an Excel or CSV file. Connect the AI you already use. Ask in plain words and get an answer that was computed, not guessed.

Start free See how it works

No card. Sign in with Google, Microsoft, GitHub or an email. Free files are kept for 24 hours.

A penguin in a small office slides a spreadsheet across the desk to a friendly robot, which hands back a single neat answer card

Three steps, about five minutes

Every sheet in your file becomes a table with proper columns and types. From then on, SQL does the work, whether you write it or an AI does.

  1. Add a spreadsheet

    Drop an .xlsx, .csv or .parquet file on the Files page. Cells are read, never executed. You see every table it produced, with row counts, before asking anything.

  2. Ask it something

    The Query page runs your first questions right in the browser. Click a table to see its rows, then try a total, a filter, a join across two sheets.

  3. Connect your AI

    In Claude or ChatGPT, add tableski as a connector and sign in; no keys to copy. Claude Code, Cursor and kowalski take a ready-made setup. Then ask in plain words: the AI writes the SQL, tableski runs it on your file.

Which customers are more than 60 days late, and how much do they owe in total?

Four customers, €18,420 in total. Largest is Northwind Traders at €9,800, 74 days.

The AI turned that into one query over your invoices sheet and ran it here:

SELECT customer, sum(amount) AS owed, max(days_late) AS days
FROM invoices
WHERE days_late > 60
GROUP BY customer
ORDER BY owed DESC
customeroweddays
Northwind Traders9,800.0074
Blue Harbour Ltd4,120.0091
Kestrel & Sons2,900.0063
Marlow Foods1,600.0068

Numbers come from your cells. If a total looks wrong, the SQL is right there to check.

Two kinds of tables

The same engine, two sources of data.

Your files

For anyone who lives in spreadsheets: sales, invoices, stock, budgets, exports from the systems that will not talk to each other.

  • Excel workbooks, CSV, Parquet, NDJSON; one table per sheet
  • Joins across sheets and files, dates that are dates, typed numbers
  • Results exported back to .xlsx or .csv
Start free

Live market data coming

For traders: prices and indicators as tables the AI can query next to your own positions sheet. "Which of my holdings dropped below their 50-day average this week?"

  • Minute, hourly and daily bars for a launch list of symbols
  • Indicators as SQL: moving averages, RSI, MACD, Bollinger, drawdown
  • Joined with your uploaded portfolio, in one query

Why it works

Computed, not guessed

The AI never reads your cells into its memory and adds them up in its head. It writes SQL; a real query engine computes the answer over every row.

Fast enough to talk to

A 100,000-row workbook answers in tens of milliseconds, a million-row CSV in well under a second. Measured, with numbers.

Read-only by construction

Only SELECT-shaped queries over your own tables run. Timeouts, row and memory caps, per-minute limits. How the guard works.

Your file stays yours

Nothing is written back into your workbook. Free files are removed after 24 hours; delete anything, or the whole account, at any time. Privacy.

Plans

Start free and stay free for one file at a time. Pay only when you want files kept and more of them.

Free

€0
  • 1 file, up to 1 MB
  • 100,000 rows per file
  • 25 questions a day
  • Files kept for 24 hours

Starter

€9 a month
  • 5 files, up to 10 MB each
  • 250,000 rows per file
  • 500 questions a day
  • Files kept until you delete them

Pro

€29 a month
  • 20 files, up to 50 MB each
  • 2 million rows per file
  • 2,000 questions a day
  • Files kept until you delete them

Beta prices; they may change with 30 days' notice. Payments are handled by our merchant of record, taxes included where they apply.

Questions people ask first

Do I need to know SQL?

No. Once an AI is connected you ask in plain words and it writes the SQL. The Query page is there for people who like to see or tweak the query, and every answer shows the query it came from.

Which AI assistants work?

Claude on the web, desktop and phone, and ChatGPT: add tableski as a connector and sign in. Claude Code, Cursor, kowalski and any other MCP client: paste a ready-made setup from the Connect page.

What files can I upload?

Excel workbooks (.xlsx, .xls), OpenDocument (.ods), CSV, Parquet and JSON lines. Each sheet becomes a table. Formulas are read as their last calculated values; charts and macros are ignored.

Can the AI change or delete my data?

No. Only read-only queries run, over your own tables, with time and size limits. Your workbook is never written to; exports go to new files.

Who can see my files?

You, and the AI apps you connect, which you can disconnect any time on the Tokens page. We do not look at file contents except to fix a fault you report. Free files are deleted after 24 hours; you can delete anything, or the whole account, whenever you like. Details in the privacy policy.

What does it cost, and how do I cancel?

The free plan costs nothing and needs no card. Paid plans are monthly, with tax included where it applies, and you cancel in the billing portal linked from the Usage page; the plan stays until the end of the month you paid for.

Or run it yourself

The engine is open source, one binary, MIT or Apache-2.0. Point it at a file and connect any MCP client.

curl -fsSL https://raw.githubusercontent.com/yarenty/tableski/main/install.sh | sh
tableski --file your-workbook.xlsx
claude mcp add --transport http tableski http://127.0.0.1:8080/

GitHubcrates.ioBinariesDocsFive-minute demo