Numbers, read carefully.

Reconciliations, budgets, a spreadsheet nobody has understood since the person who built it left. It reads the numbers and shows the working - and stops short of telling you what to buy.

Three steps.

  1. 1

    Attach both sides

    The bank export and the ledger, or the budget and the actuals.

  2. 2

    Ask for the difference

    Not the summary - the rows that do not match, with the amount.

  3. 3

    Check the working

    It shows how each figure was reached, which is the part you can argue with.

A live chart of the Nestle share over the last year, with a switch for the period above the line.
In the app. A live chart of the Nestle share over the last year, with a switch for the period above the line.

Prices and figures, drawn.

Ask about a share, a coin or an index and the answer carries a live chart with today's price. Figures from your own files become key numbers and charts.

What it does.

  • Reads a spreadsheet including its formulas and says what a cell actually computes
  • Reconciles two lists row by row and names what is only on one side, where an amount differs and where one side has duplicates
  • Works out loan payments, amortisation, present value and internal rate of return, rather than estimating them
  • Checks an IBAN, a CHE number or an ESR reference before it goes on a payment
  • Computes the statistics on a series - median, quartiles, outliers - instead of describing them
  • Builds a model with the assumptions written out, so you can argue with them
  • Charts a series in the conversation and says what changed and where
  • Draws a live chart for a share, a coin or an index, with the real price and a period you can switch
  • Sets a price alert from the chat, like “tell me when Nestlé falls below 70”, and emails you when it is crossed
  • Drafts the commentary that goes with the figures, in the register the audience expects

Where it stops

No investment advice. Not because of a disclaimer, but because a system that produces plausible text has no business telling anyone where to put their money. Arithmetic in a language model is reasoning, not calculation: for anything that has to add up, have it show the working and check the total.

Asked like this

Two exports, one from the bank and one from the ledger. Find the entries that appear in one and not the other, and group them by likely cause.

Which size to use

Large. Arithmetic across many rows is where the small models slip quietly.

Compare the sizes

Questions.

Will it tell me what to buy?

No. A system that produces plausible text has no business advising on money, and saying so is not a disclaimer - it is the reason the feature does not exist.

Can it read formulas?

Yes. It reads what a cell computes, not only what it displays - which is usually where the surprise is.

Can it watch a price for me?

Yes. Say “tell me when Nestlé falls below 70” and you get an email once the price crosses that line. Prices can be delayed, and an alert is a reminder you set, not a recommendation.

Is my data used for anything else?

No training, no analysis, no profiling. Your figures are processed to answer your question and for nothing else.

Other kinds of work.

Companies

Own account, own keys, prepaid credit, no seat count.

Public sector

Swiss operator, named subprocessors, nothing used for training.

Sensitive data

What is stored, what is not, and what you can switch off.