When trying to parse a PDF containing tabular data the Kagi Assistant replies with "We are sorry, we are not able to extract the source". This occurs with different AI backends selected, so I suspect it's an issue with the PDF pre-processing done by Kagi.
While I can't provide the original document (personal finances!), I've managed to build a test pdf which fails in the same way. Note it isn't one specific pdf file that fails, it's all pdf files generated by a specific source.
The prompt "What does this pdf say?" with this file:
fails.
The following libraries successfully parse text out of my original file as well as my test file:
- pdfminer.six (python)
- PyMuPDF (python)
- PdfPig (c#)
pypdf fails, and my test file produces a similar (but not identical) error to my original file relating to "/Subtype".
Kagi Assistant should support parsing of PDFs that are supported by popular PDF parsing libraries and viewers.
