Extract header fields
Reads the supplier, total and due date. Each value links to its place on the page.
supplierNovák Stavby s.r.o.
total18,330.00 EUR
due_date2024-05-12
How an invoice moves through a workflow
The first node extracts the supplier, total and due date. A filter sends invoices over 10,000 EUR to a second node that extracts the line items. Every other invoice is returned straight away.
Filter large invoices
Checks the total. Invoices over 10,000 EUR go on to the next node.
totalis greater than10,000 EUR
This invoice18,330.00 EUR
yesContinues to Line items
noReturns straight away
Extract line items
Reads every row of the item table.
| item | qty | price |
|---|---|---|
| Concrete C25/30 | 18 | 980.00 |
| Delivery | 1 | 690.00 |
| Sum of rows | 18,330.00 EUR | |
Your application gets one JSON document
The response holds every value the workflow extracted. The same body can be sent to a webhook.
POST /v1/workflows/incoming-invoices/runs{
"workflow": "incoming-invoices",
"supplier": "Novák Stavby s.r.o.",
"total": { "amount": 18330.00, "currency": "EUR" },
"due_date": "2024-05-12",
"line_items": [
{ "item": "Concrete", "qty": 18, "price": 980.00 },
{ "item": "Delivery", "qty": 1, "price": 690.00 }
]
}Why Pactus API
Why choose Pactus instead of a language model API?
One endpoint to integrate
You send a document and get structured data back. We handle the models, configuration, output format and updates.
Visual workflow builder
Easily organize your data extraction pipeline in a visual workflow editor.
Testing before integration
Play around with your workflow on your own documents and see the result before you write any code. When you change the schema, you see the effect right away.
Cost known in advance
You pay with prepaid credits, and every test run shows how many credits a document uses. You know what a thousand documents will cost before you process them.
Reliable output
We optimize the data extraction for cost, accuracy and adherence to your JSON schema. We use our own OCR models to read even poor scans and handwriting reliably.
Traceable values
Every extracted value comes with its page number and position in the document. You can check any value against the original document.
Pricing
You pay a fixed price per extracted page. The price is adjusted by the number of fields and by your choice of language model.
Credits per page
Here's how we calculate the price per page.
page=OCR+model×(1 + 0.2 × fields)
- OCR is charged only for scanned pages. Pages that were already read are not charged again.
- model is the price of the language model you choose.
- fields is the number of fields you extract. Every field adds 20% to the price.
Example
A 2-page invoice goes through OCR and three values are extracted, "customer", "date" and "price".
OCR$0.001/page
Extraction, 3 fields*$0.002 × (1 + 0.2 × 3) = $0.0032
Per page$0.0042
2 pages$0.0084
* Price with GPT-5.6-luna as the extraction model.
Enterprise
For companies whose documents have to stay on their own infrastructure.
- Deployment on your own infrastructure
- Runs on a local open-weight language model
- Annual license instead of per-page pricing
- Documents never leave your network
- Dedicated support
Build your first workflow
Sign up, upload a few example documents and see Pactus in action.