OmniDoc Document Extraction

Document structure extraction

Get the headings, paragraphs, tables and images of a PDF, a scan or an Office file in the order and hierarchy of the original, as Markdown and JSON you can use right away.

Start extracting for freeDesktop app: no page limit

Up to 21 pages per PDF and no monthly cap; free jobs wait in a queue. For longer files or when you are in a hurry, the OmniDoc desktop app has no page limit and no queue.

How it works

  1. Upload a file

    Sign in with an OmniDoc account and drop the file on the page.

  2. Extract

    PDFs, scans and images go through layout analysis, text recognition and table recognition; Office files, e-books and web pages are parsed directly.

  3. Download

    Download a ZIP with Markdown, JSON and the images. The job runs in the background, so you can close the page.

What the result keeps

Heading levels

Chapters, paragraphs and lists keep the hierarchy of the original.

Rows and columns

Tables come out as Markdown tables and structured data, not as a run of numbers.

Scans

Scanned PDFs and images are recognised first, then organised by layout.

Images

Pictures are saved in images/ and referenced from the Markdown and JSON.

What is in the ZIP

document.mdThe full text as Markdown
document.jsonStructured data in reading order
images/The pictures of the document, saved separately
manifest.jsonA list of the files with basic information

Supported formats

FileHow it is processed
PDF, scanned PDF, PNG, JPG, WEBP and BMP imagesLayout analysis, text recognition (OCR) and table recognition
Word, PowerPoint, Excel, CSVThe document structure is read directly
EPUB e-books, HTML pagesChapters and body text are parsed
Markdown, TXT, SRT and ASS subtitlesNormalised into the same Markdown and JSON

Example

Page 1 of an English report with a title, data cards and a city table
Page 1 of the original PDF
# Greener Last-Mile Delivery
How five cities cut delivery emissions by a third in one year

## 1 What changed
In 2026 the five pilot cities replaced diesel vans on their busiest routes with electric vans and cargo bikes, …

**Table 1** Results by city, January–September 2026
| City | Parcels per day | Electric share | CO₂ per parcel (g) |
|---|---|---|---|
| Hangzhou | 182,000 | 74% | 96 |
| Shenzhen | 236,000 | 81% | 88 |
| Chengdu | 141,000 | 63% | 112 |
The extracted document.md (excerpt)

Questions

How is this different from copying the text?

Copying gives you plain text. Extraction keeps the headings, paragraphs, tables and images and how they relate, as structured data a program can read.

Is the recognition accurate?

PDFs with a text layer and Office files are read directly. Scans and images are recognised; check numbers and tables against the original.

Do I need an account?

Yes. Sign in with an OmniDoc account; signing up is free. Only you can see the files you upload and their results.

What are the limits of the free version?

Up to 21 pages per PDF and no monthly cap; free jobs wait in a queue. For longer files or when you are in a hurry, the OmniDoc desktop app has no page limit and no queue.

Do I need to install anything?

No, it runs in the browser. For long files, or when you are in a hurry, use the OmniDoc desktop app for Windows, macOS or Linux.

Try it now

Up to 21 pages per PDF and no monthly cap; free jobs wait in a queue. For longer files or when you are in a hurry, the OmniDoc desktop app has no page limit and no queue.

Start extracting for freeGet the desktop app