Upload and share
Supported files
Which file types Sergo reads, and what Basic and Standard (AI) get out of each.
You can upload any file up to the size limit, but only some types can be processed into search. This page describes the two presets a fresh install offers. An admin can change them, so your install may differ; the processing dialog shows which of your files each preset can read.
| File | Basic | Standard (AI) |
|---|---|---|
| Reads the text layer | A model reads each page | |
| Word (.docx) | Reads the file directly | A model reads each page of a PDF copy |
| PowerPoint (.pptx) | Reads each slide directly | A model reads each page of a PDF copy |
| Other Office files: .xlsx, .doc, .ppt, .xls, .odt, .odp, .ods | Reads the text layer of a PDF copy | A model reads each page of a PDF copy |
| Text (.txt, .md) | Reads the text as is | Reads the text as is, for free |
Other types, such as images and .hwp files, show Can’t process and stay out of search.
Basic is fast and free, but it only gets text that is already in the file. Standard (AI) sends each page to a language model, which also reads scans, tables, charts and slides, and costs credits. See Processing.
Basic reads the text layer. It has no OCR, so:
- A scanned PDF, with no text layer at all, comes out as No text, with a note that Standard (AI) reads scans.
- If only some pages have no text, the file is processed with a warning such as “3 of 40 pages have no text”.
- Some PDFs have a text layer that turns into nonsense characters when it is copied, because their fonts don’t say which character each shape is. Basic notices this. If every page is like that, the file comes out as No text; if only some are, it keeps the text and adds a warning.
Standard (AI) reads every page, scanned or not. It sends pages with a broken text layer as a picture, so the model reads what the page shows.
A PDF that is damaged or protected with a password fails with a note saying so.
Basic reads the .docx file itself, not a PDF of it, so headings and tables keep their structure. It reads:
- paragraphs, headings, and numbered lists with the numbers Word shows, such as “4.2”;
- tables, including tables inside table cells;
- text boxes, content controls, and text added with tracked changes;
- footnotes and endnotes, right after the paragraph that refers to them;
- headers and footers, once each, in a section at the end.
It skips the table of contents, and does not read comments or text inside pictures. A Word file with no text at all comes out as No text.
A Word file has no fixed pages: where a page ends depends on the program that draws it. So the
page numbers start as estimates, shown with ≈ in search results and in the document view. Sergo
makes a PDF copy in the background and then gives the chunks the pages of that copy.
The older .doc format goes through a PDF copy, like the other Office files.
PowerPoint
Section titled “PowerPoint”Basic reads the .pptx file itself. Each slide is one page, and hidden slides keep their place, so page 12 is always slide 12. On each slide it reads:
- the title first, then the other text from top to bottom and left to right;
- lists, tables, and text inside grouped shapes;
- charts, as a table of their values;
- SmartArt;
- text in pictures pasted as Windows metafiles, such as a pasted Excel table;
- speaker notes, after the slide’s text, marked “Notes:”.
It does not read the alt text of pictures, or text that is only in a picture. Basic also makes a PDF copy in the background, for viewing the slides.
A file saved as Strict Open XML can’t be read by Basic. Save it again as a normal .pptx, or use Standard (AI).
Standard (AI) reads hidden slides too, so they count in the estimate and the charge.
Other Office files
Section titled “Other Office files”Spreadsheets, the older .doc, .ppt and .xls formats, and OpenDocument files are converted to PDF first. Basic reads the text of that PDF. That works for running text, but table cells come out without their order, and the values of charts are often not in the text at all. Standard (AI) reads each page of the PDF as it looks.
Plain text and Markdown are read as they are, by either preset, for free. They have no pages, so their page numbers are estimates.
PDF copies
Section titled “PDF copies”Sergo converts an Office file to PDF once, with LibreOffice, and keeps the copy. Processing the file again, with any preset, reuses it. The page numbers of chunks read from a copy are the pages of that copy, and the document view shows the copy because browsers can’t show Office files.