Something else

What to hand Canonry

Whatever your tool can export: one file, or a zip of a folder of mixed files nested as deep as it likes. If Canonry does not recognise the shape of what you upload as one of the sources above, it runs the generic playbook automatically rather than stopping.

The upload box suggests the extensions the guides ask for, and that is a hint rather than a rule: Canonry reads what a file is from its bytes, so a file with an unusual extension or none at all is still read as text if that is what it holds. If your file manager will not let you choose it, switch the dialog to all files, or zip it.

What counts as a document

Every file Canonry can read as text becomes one document, with its own line in the estimate and its own cost: plain text, Markdown, HTML, RTF, CSV, JSON, and a PDF or a Word document sitting in the folder alongside the rest. There is no list of allowed extensions, so a file your tool gave an unusual name, or no extension at all, still counts as long as its contents are text.

Four kinds of file are skipped instead, and the document count on the estimate screen is where you see it. An image, a sound or video file, an archive, or anything else whose bytes are not text: handing those to a model as if they were text is worse than leaving them out. Anything hidden, meaning a file or folder whose name starts with a dot, which is where an application keeps its own bookkeeping rather than your notes. An empty file. And a scanned PDF with no text layer, because in a mixed folder there is nothing to read out of it: upload that PDF on its own and Canonry recognises it as a PDF instead, and looks at each page as an image.

What to expect

The generic playbook reads what it can and proposes entities from whatever text and images it finds. An image is never a document of its own, but the playbook can store one it finds referenced from a document it is reading and attach it to what it proposes. HTML, RTF and CSV are handed over as their own text, markup and commas included, rather than converted to something tidier first.

It is slower and looser than a playbook built for one source, because it is not assuming a particular structure, but it is a legitimate result, not a fallback that failed. If your source is one of the named ones, use its guide instead: a named playbook understands its structure and produces better candidates from it. You can also pick a named playbook yourself on the confirmation screen when Canonry guessed generic and you know better.