Reading PDFs

Ask DataMind Oda about a PDF and it reads the file itself, page by page, so it can answer from the document's own text. Scanned pages arrive as page images when Oda can view them. This page covers what a read returns, how Oda moves through a long document, and the limits it works within.

Ask Oda to read a PDF

1. Point Oda at the file in your message, for example Summarise annual-report.pdf or What does the contract in this folder say about renewals?
2. Oda reads the PDF on your computer and answers from its contents. The chat shows a Read step for the file; Reading files and images covers what the step displays.
3. For a longer document, Oda reads it in windows and continues from where each window stopped.

You do not have to pick pages yourself. To read a specific part, say so: Read pages 12 to 15 and tell me the payment terms.

What a read returns

Each read opens with a line naming the document, like PDF: 25 pages, title "Annual report", and gives the text page by page. Every page starts under its own marker, --- Page <n> ---:

• The text keeps the layout of the page: Oda rebuilds lines and paragraphs from where the words sit on it.
• Aligned cells come through as table rows, with a | between them.
• Text comes from the document's own text layer. Oda does not run OCR, so a page that is only a picture has no text to give (see Scanned pages).

Work through a long document

• By default, one read takes up to 10 pages: the first 10 of the document, or all of it when it is shorter.
• A read that names a page window can carry up to 20 pages. A larger request comes back as pages selects <n> pages; read at most 20 pages per call.
• Each window ends with a footer that marks where to continue, for example (Showing pages 1-10 of 25. Use pages="11-20" to continue.). When the read reached the end, the footer reads (End of PDF - total 25 pages).
• Even a window that fits the page limit can stop on the text it carries: (Output capped. Showing pages 1-2 of 4. Use pages="3-4" to continue.). The remaining pages arrive on the continuation.

Read only the pages you need

Name the pages in your message and Oda passes the selection to the read. A selection takes a single page, a range, or a mix: 3, 1-5, 2,4,7-9.

• Pages are numbered from 1, and a selection is read in page order with duplicates taken out.
• A selection that cannot be used comes back with the reason, for example pages are numbered from 1, pages range "<part>" runs backwards; write it as <last>-<first>, or page <n> is out of range; the PDF has <m> pages.
• Page windows belong to PDFs: pages applies only to PDF files; use offset and limit for text files.

Scanned pages

• A page with no extractable text but something drawn on it counts as a scan. Oda shows such a page as an image right after its text, marked like [page 2: no extractable text, likely a scan; shown as image 1 after this text]. A page with only a little text is handled the same way: [page 1: little extractable text; shown as image 1 after this text].
• A page that has neither text nor an image shows [page <n>: no text].
• A page image is shown only when Oda can view images in this chat and the read's limits allow it. When a scan cannot be shown, the read says so in place, for example (Scanned pages are not shown as images: the current model does not accept images.); the note names the reason, such as at most <n> page image(s) fit in one result.

Limits and refusals

• Size. A PDF over 100 MB is refused: cannot read "<path>": the PDF is <size> MB, above the 100.0 MB limit for reading PDFs; split it or extract the pages you need first. Split the document, or make a copy with only the pages you need.
• Time. A read stops after 120 seconds with the PDF took longer than 120 s to read. Try again with a smaller range of pages.
• Memory. A document too heavy for the reader stops with the PDF needs more memory than the reader allows.
• Reader failure. the PDF reader failed (<message>) or the PDF reader stopped unexpectedly (exit code <code>).
• Password. the PDF is password-protected; ask for a copy without a password.
• Damage. the file is not a valid PDF or is damaged (<detail>), or the PDF could not be parsed (<detail>).
• At once. Oda reads up to two PDFs at a time; further reads wait their turn.
• Reading never changes the file. A read does count as having read it, so if Oda replaces that file later the usual read-before-replace checks apply (Read, write and edit files).

Questions, answered

How many pages does one read cover?

Up to 10 pages by default: the first 10 of the document, or all of it when it is shorter. A read that names a page window carries up to 20 pages; a larger request comes back as pages selects <n> pages; read at most 20 pages per call. Longer documents are read in windows.

How do I ask for specific pages?

Say which pages you want in your message, for example read pages 12 to 15. Oda passes the selection to the read: a single page (3), a range (1-5), or a mix (2,4,7-9). Pages are numbered from 1, and the selection is read in page order.

Why did the read stop before the end of my document?

Two reasons. A default read takes the first 10 pages, so a longer document continues in windows; and a window can stop early when the text it carries hits the read's output limit, marked (Output capped. Showing pages 1-2 of 4. Use pages="3-4" to continue.). Either way the footer says where to continue, and Oda continues from there; you can also ask for the next pages directly.

What happens to scanned pages?

A scan has no text layer, so there is no text to extract. When the page can be shown as an image, Oda receives it as one and can answer from what the page shows; the read marks it, for example [page 2: no extractable text, likely a scan; shown as image 1 after this text].

Why are scanned pages not shown as images?

The read names the reason in place, for example (Scanned pages are not shown as images: the current model does not accept images.). Common reasons: the model in this chat does not accept images, or the read can fit only a limited number of page images in one result. The markers stay in the text either way.

The text Oda read looks different from the document. Why?

Oda rebuilds lines and paragraphs from where the words sit on the page, so unusual layouts can come through differently than they look. Aligned cells come through as table rows with a | between them, and text that is part of a picture is not extracted; those pages arrive as images instead.

Why can't Oda read my PDF?

Some PDFs cannot be read. A password-protected file comes back as the PDF is password-protected; ask for a copy without a password, and a file that is damaged or not a valid PDF as the file is not a valid PDF or is damaged (<detail>) or the PDF could not be parsed (<detail>). A copy without the password, or a fresh export of the file, can be read.

Why did the PDF read time out?

A read that takes longer than 120 seconds stops with the PDF took longer than 120 s to read. Try again with a smaller range of pages, or split the document.

My PDF is too large. What can I do?

A PDF over 100 MB is refused: cannot read "<path>": the PDF is <size> MB, above the 100.0 MB limit for reading PDFs; split it or extract the pages you need first. Split the document, or make a copy with only the pages you need, and read that.

Can I read a PDF in Oda without asking a question about it?

Yes. Open the file in the preview and read it there; Preview PDFs and Office files covers the page view, selection and zoom.

Does reading a PDF change the file?

No. A read only looks at the file. It does count as having read it, so if Oda replaces that file later the usual checks apply; they are on Read, write and edit files.