Add Abacus.AI

[carstenschroeder]

This is a WIP.

Abacus.AI only provides the date and amount in the web portal, not a unique invoice number.
If both fields are the same for two invoices, the second invoice is ignored. But worse still, the following invoices are downloaded out of order because the table row is not also ignored. This means that the PDF for October 2025 is downloaded for November 2025.

So far, I have gotten around this by making the field attribute tuples unique via a random field and extending the invoice number by the index. But this is obviously a dirty hack.

It would be great if you could create a way to extract the invoice number from the PDF in such cases.

[tobiaslins]

We where thinking about adding a way to use the PDF hash as ID for providers that don't generate PDFs on the fly.
Can you please check if you download one invoice twice, if they have the same md5 hash? Thanks!

[carstenschroeder]

Unfortunately the hashes are not equal. The PDFs are created on the fly.

PS C:\Users\Carsten\Downloads> Get-FileHash -Path ".\da5a1f13c_invoice (1).pdf" -Algorithm MD5

Algorithm       Hash                                                                   Path
---------       ----                                                                   ----
MD5             F84AF8306CDEED4BD1F3C70C60E20912                                       C:\Users\Carsten\Downloads\da...


PS C:\Users\Carsten\Downloads> Get-FileHash -Path ".\da5a1f13c_invoice.pdf" -Algorithm MD5

Algorithm       Hash                                                                   Path
---------       ----                                                                   ----
MD5             2C3503F6FF9E8B78B7D0E1CCB28F1B09                                       C:\Users\Carsten\Downloads\da...

[carstenschroeder]

@tobiaslins
I worked around the issue with the duplicate dates by defining a variable called “random,” which is populated with the date plus a random number for each row. That works.

But could it be that Invoice Radar has a bug in how it indexes table rows in this specific case? In our example for abacus.ai, the table containing the invoices has 16 rows. The first 8 rows are omitted because the invoice dates are prior to the earliest month (12/2025). Of the remaining invoices, the first 5 have already been loaded. However, invoices 14–16 are not downloaded; instead, invoices 6–8 are downloaded (but with the metadata of 14-16).

Even in the Developer Tools, rows 6–8 are displayed as relevant, which is incorrect.

Could you please help? I'm stuck since months.