CSV Is an Interface Your AI Can Read

If you use Claude or ChatGPT for business questions, you have probably done one of two things. You pasted a screenshot of a dashboard, or you connected your whole account and hoped for the best. Both work, sort of. Both have a cost you only notice later.

There is a third option, and it is older than both. A plain CSV file with a date in its name. It is not glamorous. It is the best AI interface you already have.

The Three Ways to Show an AI Your Numbers

Screenshots are lossy. The AI reads pixels, guesses at truncated labels, and cannot sum a column it cannot see. You get a plausible answer built on a partial picture. When the answer is wrong, you cannot tell, because you also only saw the picture.

Connecting an account is the opposite problem. The integration reads everything, forever. Every customer, every order, every email, on someone else's schedule, under someone else's terms. You wanted an answer about last month's orders. You granted a standing subscription to your entire business.

A CSV sits in between, and that is exactly where you want to be. It is complete where a screenshot is partial. It is bounded where a connected account is not. You choose the file. The AI sees the file. Nothing else.

Why the Dated CSV Wins

A CSV has four properties that matter when an AI is on the other end.

• Plain. Rows and columns, no rendering, no guessing. An AI can sum, filter, and compare the actual values.
• Portable. The same file works in Claude, ChatGPT, Excel, or a Python script. No adapter, no plugin.
• Inspectable. You can open it and read exactly what you are about to share, line by line, before you share it.
• Scoped. One file is one slice. Last month's orders is last month's orders. Not your customer list, not your inbox.

The date in the filename adds a fifth property. History. A folder of dated snapshots is a record of how your numbers moved, and 'compare these two files' is a question any AI answers well.

A Worked Example

Say you run a Shopify store and something feels off about margins. Take last month's orders snapshot, one CSV, and drop it into an AI chat. Ask: 'Group these order lines by product. For each product, show units sold, average selling price, and average discount. Which product's realised price is slipping compared to the first half of the month?'

The AI does the arithmetic on real rows, not on a picture of rows. It might tell you one product's average discount doubled in week three, which is when you launched that promotion and forgot to end it. That is a concrete, checkable answer. You can open the same file and verify every number it used. And the AI saw one month of orders, nothing more.

How Toqool Produces These Files

A Toqool connector is a set of readable Python scripts with a single dependency, requests. They run on your machine and pull your platform data into dated CSV snapshots on your own disk. Each run adds a new snapshot next to the old ones, so history builds up locally without you doing anything special. A one-click run script is included, and an optional daily scheduled pull if you want the folder to fill itself.

The toolkit is read-only by design, and it prints that posture when it runs. Before any pull, verify_auth.py checks your credentials and gives a precise diagnosis per failure, so you are never guessing whether the problem is a token, a scope, or the platform. Setup is AI-guided via an included runbook. For shop platforms it is quick. For Microsoft 365, budget about 45 minutes including admin consent.

This is why we call the CSVs AI-legible. The columns are clean, the names are plain, the dates are in the filenames. You can hand a single file to Claude, ChatGPT, or Excel and share only the slice a question needs. The safest data is the data that never left, and with this setup, the rest of it never does.

The AI industry keeps building richer pipes into your business. You do not need a richer pipe. You need a clean file, a date in its name, and the choice of when to hand it over. That interface has existed for decades. Now it finally has a reader worthy of it.