Download free

A document into a table, with the doubtful rows marked

Drop a catalog, an invoice, a price list, a bank statement, a scanned PDF or a photo, and get back a spreadsheet where the rows it is not sure of are marked «review», each with the reason.

Updated · version 2.0.242 · Markdown

Copying a catalog or an invoice into a table is slow, and the dangerous part is the figures: one comma read as a decimal point and an order goes out ten times too big. Your Nobreinier does it for you and, instead of guessing, tells you which rows to check.

How to ask

Drop the file in the chat (or tell it where it is in the project) and say what you want:

  • «Pass this catalog to a spreadsheet.»
  • «Put this invoice in a table and tell me what does not add up.»
  • «Make a CSV of this bank statement.»

It works with PDFs (with text or scanned), photos, Word, spreadsheet and OpenDocument files, and CSV. Spanish and English documents are read the same way.

What you get back

  • A spreadsheet (.xlsx) or a CSV, in the «tables» folder of the project, with the document's columns and three more at the end: Review, Reason and Source (page and line of the original).
  • A line like «42 rows, 3 to review».
  • A second sheet, «Checks», with how the document was read and each of its own totals: what it says and what comes out.

The rows to review are shaded, and the exact cell that raised the doubt is shaded darker. In that cell you see what the document said, never a corrected figure.

What is checked

Every figure is checked against the document itself, not against what a program «thinks» it should be:

  • Quantity × price against each line's amount (with its discount, if it has one).
  • The lines against the subtotal, the parts against the total, and a tax against its own rate.
  • In a statement, each row's balance against the previous balance plus the movement, and the dates in order.
  • Prices of one code that appear twice with different values.
  • Dates and currencies written in a different way from the rest of the document.
  • On a photo or a scan, the words read with low certainty.

A figure that cannot be read at all (a letter «O» instead of a zero, for example) is kept as it was read and marked.

Commas and points

«1.234,56» and «1,234.56» are both read correctly: the program looks at which separator the document uses in the figures that can only mean one thing (for example «12,50») and applies it to the whole document. If the document gives no clue, for example a list of prices like «1,250» and «7,490», it does not guess: it leaves those figures as they are, marks them and asks you once whether the comma marks thousands or decimals. Your answer settles the whole document.

Dates work the same way: «15/03/2026» shows that the day comes first. If nothing in the document says so and the dates could be read both ways, they are marked.

What it does not do

  • It does not correct anything: a row marked «review» is for you to compare with the original.
  • It does not guess a missing number or a tax rate.
  • A blurry photo can fail to read some figures; they will show up as marked rows. A photo taken at an angle is straightened by the outline of the page, so keep the whole page in view on a table or background that contrasts with it; if no outline is found, the photo is read as it is and the answer says so.
  • A scan or a photo is read as printed: a cell whose text is cut off in the picture comes out cut off.
  • A document with no table (a letter, for example) gets no table: it says so.