document in.
api out.
Upload a sample, name the fields you need, test on real files, publish an endpoint. patrapi reads PAN cards, sanction letters and light bills so your software doesn't have to.
POST /v1/workflows/wf_sanction/extractionsDate: 12/08/2026
To, M/s Shree Ambica Textiles, Surat
We are pleased to sanction a Term Loan of ₹ 48,50,000/- for purchase of machinery.
Repayment: 84 monthly instalments after a moratorium of 6 months.
Rate of interest: RLLR + 1.50% = 9.75% p.a.
Margin: 25%. Covered under CGTMSE: Yes.
Simplified. Each field also returns its clean value, the page, the source text and where on the page it sits.
sample file to live endpoint, in one sitting.
No annotation, no training run, no onboarding call. If you can describe a field to a new intern, you can build the workflow.
drop in a sample
A scan, a phone photo, a 40-page PDF. English, Hindi, Gujarati, or all three.
name your fields
Type them in plain words, or let the assistant read the sample and propose them.
test on real files
See each value next to the text it came from. Fix what's off, run again.
publish the api
A versioned endpoint and a key. Change the workflow later without breaking callers.
not sure what to extract? ask.
It reads your sample, suggests fields with types and rules, and tests them on your files before you publish anything.
every field shows its work.
The value as printed, the clean version, the page, and the exact text it came from. When something isn't on the document, patrapi says so instead of making it up.
"loan_amount": {
"value": "₹ 48,50,000/-",
"normalized": 4850000,
"status": "present",
"page": 1,
"source_text": "sanction a Term Loan of ₹ 48,50,000/-",
"bbox": [0.12, 0.41, 0.88, 0.45]
}- present
- found, with the source text to prove it.
- not_present
- not on this document. null, never invented.
- unreadable
- it's there, but the scan won't give it up.
- conflicting
- two different values on the page. flagged for a human.
works on any document.
starts with india's.
patrapi reads whatever you point it at. But nobody should start from a blank page, so we wrote recipes for the paperwork behind every MSME loan and machinery subsidy file.
A recipe is just a workflow someone already built. Use it as is, or fork it, add the field your bank asks for, and publish. It's yours.
identity
05PAN card, Aadhaar card, Birth certificate, School leaving certificate, ST / SC / OBC caste certificate.
business
09Udyam Registration Certificate, GST registration certificate, PAN card of a firm or company, Registrar of Firms (ROF) certificate, Certificate of Incorporation, Memorandum and Articles of Association, Partnership deed, Gujarat Pollution Control Board consent, MANTRA certificate.
property
07Rent / leave and licence agreement, Property tax bill, Index copy, Sale deed, Dastavej, Approved building plan copy, Construction permission letter.
power & utilities
09Electricity bill, Proforma-15, Electricity connection quotation, Power estimate copy, Power released letter, HT checking sheet, Electrical inspector report, GEDA certificate, Tri-party agreement for solar installation.
finance
08Income tax return, Sale bill / tax invoice, Indigenous machinery purchase bill, Machinery payment receipt, Cancelled cheque, CC / OD / CA or bank account statement showing rent paid, Loan account statement, Loan sanction letter.
your document
+Not on the list? Upload one and build it in less time than it takes to read this page.
- loan & subsidy consultants
- every field an application needs, from the client's file into your tracker.
- ca & tax practices
- ITRs, GST certificates, bank statements and invoices, read before you open them.
- lenders & nbfcs
- KYC and business documents into your loan system, with evidence behind each value.
- any team with a form
- onboarding, claims, vendor checks. if someone retypes documents today, it's a workflow.
no training. no setup fee.
no sales call.
There's nothing to train on your documents. Describe the fields, test on a few files, publish. And as the AI underneath improves, your workflows get better without you touching them.
| compared on | the usual way | patrapi |
|---|---|---|
| getting started | label hundreds of samples, wait for training | one sample, a list of fields |
| adding a field | relabel, retrain | type it, test it, publish |
| when ai gets better | your model stays where it was | you get the upgrade |
| paying for it | setup fees, annual contracts, a sales call | per document, from your first credit |
two calls.
that's the integration.
Post a file to your workflow, poll for the result or wait inline. OpenAPI spec, idempotent uploads: send the same file twice, pay once.
- versioned workflows, edit without breaking production
- batches with excel export for non-developers
- your own metadata, returned on every result
curl https://api.patrapi.com/v1/workflows/wf_sanction/extractions \
-H "Authorization: Bearer $PATRAPI_KEY" \
-F file=@sanction-letter.pdf \
-F metadata='{"application_id":"APP-2291"}'curl https://api.patrapi.com/v1/extractions/ext_8Kq2… \
-H "Authorization: Bearer $PATRAPI_KEY"
{ "status": "completed",
"pages": 20,
"fields": { "loan_amount": { "normalized": 4850000, … } },
"metadata": { "application_id": "APP-2291" } }see what went through.
Each extraction lands with its pages, the fields it found and its status. Anything flagged waits in a review queue for a human to approve or reject.
| workflow | document | pages · credits | fields | status |
|---|---|---|---|---|
| sanction letter v3 | APP-2291/sanction.pdf | 20 · 4 credits | 11 / 12 | completed |
| udyam v1 | APP-2291/udyam.jpg | 1 · 1 credit | 18 / 18 | completed |
| light bill v2 | APP-2288/bill-aug.pdf | 2 · 1 credit | 8 / 9 | needs review |
| pan card | APP-2288/pan.jpg | 1 · 1 credit | 4 / 4 | processing |
not our training data.
gone after the job
Files are held encrypted only while they're read, then deleted. Never in logs or backups.
values stay out of logs
We record field names, counts and timings to run the service. Never the PAN number itself.
no surprise bills
Every result carries its page count, so you always know what a document cost before the invoice does.
pay per document.
nothing up front.
We first built this for a loan and subsidy consultant buried in paperwork. He didn't want a software project. He wanted to pay per document, only when it worked.
Most businesses want that. So that's how we price.
On sign-up, no card. Enough to run a real batch of KYC documents. Building is free too: the first 25 test and assistant runs on every workflow cost nothing (up to 100 a month).
start free1 credit is a document up to 5 pages; each further 5 pages add one. Most KYC documents are a single credit. Volume discounts as you grow.
get an api key| a document like | pages | credits | costs |
|---|---|---|---|
| PAN card | 1 | 1 | ₹5 |
| Udyam certificate | 3 | 1 | ₹5 |
| Income tax return | 8 | 2 | ₹10 |
| Loan sanction letter | 20 | 4 | ₹20 |
asked often.
does it handle hindi and gujarati?
Yes. School leaving certificates with handwritten Gujarati entries are one of our recipes. Values come back in the format you asked for.
my document isn't in the recipes.
Build it. Upload a sample, list the fields, test. Recipes are a head start, not a limit.
how are credits counted?
Per document, in bands of 5 pages: 1 credit up to 5 pages, then 1 more for each further 5. A PAN card is 1 credit; a 20-page sanction letter is 4. Every result carries its page count.
do test runs cost credits?
Not while you build. Try it and assistant runs are free for the first 25 on each workflow, up to 100 a month per account.
do i need to write code?
Not to build or test a workflow. Your software calls the API; no software? Upload a batch and download an Excel sheet.