Turn any PDF or scan into clean Markdown with one API call
The OCR API that gives your LLM, your database or your agent text it can use: headings, tables and formulas intact, from scans, photos and handwriting.
200 free credits. No credit card required.
Your Markdown or JSON shows up here.
Drop a file to see it work. 3 pages a day, no account needed.
Most OCR output still needs cleaning up
Tables come back as loose text
Rows and columns collapse into a paragraph, and your LLM has to guess which number belongs where.
Tables stay tables
GitHub-flavoured Markdown tables, or rows as arrays in JSON.
Reading order breaks on columns
Two-column pages, sidebars and footnotes get mixed into each other.
Text in the order you would read it
Headings, columns and lists are kept in sequence, page by page.
Every tool returns a different shape
Plain text from one, HTML from another, coordinates from a third.
One shape from every model
Markdown or typed JSON blocks, the same from every model.
From file to Markdown in three steps
- 1ocr_live_c6a290cbc6ec…Copy
Get an API key
Free credits on sign-up, no card.
- 2
POST /v1/ocr { "model": "auto", "file_url": "…/invoice.pdf" }POST your PDF or image
A URL, base64 or an uploaded file. Pick pages if you only need some.
- 3
# Invoice INV-2041 | Item | Qty | | Toner | 2 |
Use the Markdown
Feed it to your LLM, index it, or ask for JSON blocks instead.
200 free credits. No credit card required.
Clean text for whatever reads it next
Give your LLM documents it can read
- Headings and tables survive, so chunks keep their meaning
- Page markers let you cite the source page
- Works on scans, slides and photographed pages
## 4. Financial results | Region | 2025 | 2026 | |---|---|---| | APAC | 4.1M | 5.3M | | EMEA | 3.8M | 4.0M |
Clean output from messy documents
# Invoice | Item | Qty | |---|---| | Toner | 2 |
{ "type": "table",
"rows": [
["Item","Qty"],
["Toner","2"] ] }Markdown or JSON
Markdown for prompts and search. JSON blocks with types when your code needs structure.
Charged for the model that answered: 3 credits
It does not fail silently
If a model errors or returns an empty page, the next model in your list reads it.
The right model without thinking
Use auto and choose quality, cheapest or fastest. Or name the exact model.
Pay for pages, not seats
Credits per page, listed up front. Failed requests and cached repeats are free.
What builders say
“Placeholder quote. The problem the customer had before, and what changed after they switched.”
“Placeholder quote. A concrete number works best: pages per month, hours saved, error rate.”
“Placeholder quote. What would have happened without the product.”
Why teams choose OCRRouter
Compared with the other ways to get text out of documents.
| OCRRouter | One vendor's API | Self-hosted open source | |
|---|---|---|---|
| Models you can call | All of them, one request | One vendor's | The ones you deploy |
| Switching models | Change one word | A new integration | A new deployment |
| When a model fails | The next model answers | Your request fails | You get paged |
| Output format | Same for every model | Vendor-specific | Model-specific |
| Spend limits per key | Built in | Varies by vendor | Build it yourself |
| Time to first result | Minutes | Hours | Days |
Failed requests are free
Errors and empty results cost nothing.
Repeats are free
The same file sent to the same model is answered from cache.
Credits never expire
Top up when you need to. There is no subscription.
Files are not kept
Uploads are deleted after 24 hours. We do not train on them.
Any document in, structured text out
Questions
Not answered here? Read the docs
Free tools and guides for every document
OCR by language
Convert your first document in the next five minutes
200 free credits. No credit card required.
# Invoice INV-2041 | Item | Qty | Amount | **Total due** $319.00