Start from a file or a sample
Drop a file on the source box, click to browse, or pick a sample: a financial PDF, a scanned receipt, a multi-column paper or a contract. A chip shows which file is loaded, and you can remove it and start over.
AI Extract: layout-aware document extraction
Source document
max 100 MB · 500 pages
No extractions yet
Extracted output
No document yet
Drop a file on the left or pick a sample to see the structured output here.
Extract structure from any document
AI Extract reads a PDF, Word file, PowerPoint, spreadsheet, HTML page or image and returns its structure as Markdown, JSON or tables. Headings, reading order and table cells stay in place, so the output is ready for a language model, a search index or a script. Pick a sample in the box above or drop in your own file, choose a format and press Extract document. The size and page limit shows in the source box before you start, and the samples run without signing in.
Why use indox's AI Extract?
Choose the format and what to read
Output comes as Markdown, JSON, HTML or Tables. Switches control table structure, OCR for scans, reading order, code and formulas, and picture export. The OCR switch is marked as costing credits and code detection as slower.
Check the result, then take it with you
The output pane has Markdown, JSON and Preview tabs, with page, table, figure and parse-time counts under it. Download the file, copy it or send it to Doc Chat.
What it captures
Layout-aware, not just text
It reads the document the way a person does, keeping what plain text extraction throws away.
Table structure
Cells, headers and spans recovered into real tables.
Reading order
Multi-column pages re-sequenced into logical flow.
Headings & sections
Hierarchy kept intact for clean chunking.
Code & formulas
Plain text vs structured
Why "extract" beats "convert to text"
Same PDF. Left: what a vanilla text extractor returns. Right: what AI Extract sees.
pdftotext output, flat text, no structure
Quarterly Report 2026 Revenue by region Region Q1 Q2 Americas 4.20M 4.85M EMEA 3.10M 3.42M APAC 2.05M 2.61M Summary Total revenue grew 14.6% quarter over quarter, led by the APAC segment... chart
AI Extract output, layout-preserving Markdown
# Quarterly Report 2026 ## Revenue by region | Region | Q1 | Q2 | |----------|-------|-------| | Americas | 4.20M | 4.85M | | EMEA | 3.10M | 3.42M | | APAC | 2.05M | 2.61M | ## Summary Total revenue grew 14.6%… 
How does AI Extract work?
- 1
Add a document
Drop a PDF, Office file, HTML page or image on the source box, or pick one of the samples.
- 2
Choose format and options
Select Markdown, JSON, HTML or Tables, then switch on the table, OCR and reading-order options you need.
- 3
Extract and use the result
Press Extract document. When parsing finishes, copy or download the output, or send it to Doc Chat.
What do I get from AI Extract?
The extractor reads the layout of each page, not only its text, so tables stay tables and headings keep their level. You choose the output format, and the same extraction is available from the API for pipelines.
- Tables come back as real Markdown or JSON tables
- Multi-column pages are put back in reading order
- Headings keep their hierarchy for clean chunking
- Scanned pages go through OCR when you switch it on
- The samples run without an account
What can't AI Extract read, and what happens to my file?
Pages without a text layer. A scan or a photo has no text layer, so switch on OCR and the page is read as an image first. A very blurry or low-contrast scan can still come back incomplete.
Complicated layouts. Unusual layouts, stacked tables and handwriting can be misread. Open the output next to the source before you rely on a table or a figure.
What the output is. AI Extract returns the structure it finds. It does not summarize, correct or rewrite your document, and the output is not a certified copy.
Your file. Files are processed on indox infrastructure, are not used to train models and are removed from storage automatically. Retention depends on your plan and is listed in the limits docs.
What can I do after extracting a document?
Go beyond AI Extract: ask the extracted document questions in Doc Chat, translate it, or convert it to another format.
Developer API
Drop AI Extract into any pipeline
One endpoint. Returns Markdown or JSON. Async by default, poll for the result.
POST /docs/api/v1/extract/
curl -X POST https://indox.org/docs/api/v1/extract/ \ -H "Authorization: Bearer $INDOX_TOKEN" \ -F "[email protected]" \ -F "format=markdown" \ -F "ocr=true" # → 202 { "extract_id": "ex_8f3k…", "status": "queued" } # Poll GET /docs/api/v1/extract/ex_8f3k…/ until status=succeeded
Who can use AI Extract?
Teams building RAG pipelines
Convert document libraries into clean Markdown chunks ready for embedding.
Finance and reporting teams
Recover P&L tables and regional breakdowns from earnings PDFs.
Legal and contract teams
Preserve clause numbering, defined terms and signature blocks.
