Engineering documents
into data you can check.
Upload a spec section, a submittal or a drawing set. Jsonit returns every requirement, limit and schedule row as structured data, each with the page and quote it came from. Works on any PDF or web page; pay only for what you use.
Join 1,000+ developers who've signed up · No subscription · About $0.004 per page
Request
curl -X POST https://api.jsonit.io/v1/extract \
-H "Authorization: Bearer sk_live_your_key" \
-H "Content-Type: application/json" \
-d '{
"url": "https://news.ycombinator.com",
"schema": {
"type": "object",
"properties": {
"stories": {
"type": "array",
"items": {
"type": "object",
"properties": {
"title": { "type": "string" },
"points": { "type": "number" },
"url": { "type": "string" }
}
}
}
}
}
}'Response
{
"success": true,
"data": {
"stories": [
{
"title": "Show HN: I built a schema extraction API",
"points": 342,
"url": "https://jsonit.io"
}
]
},
"billing": {
"cost_usd": 0.0041,
"balance_usd": 9.9959
}
}Everything included
No tiers, no locked features. Results you can verify from the first request.
Every value shows its source
Each item comes back with the page and the exact words it was taken from. Jsonit checks every quote against the document and flags numbers it can’t find, so you review the few that need it instead of re-reading everything.
High accuracy when it matters
Standard mode is fast and cheap for clean documents. High accuracy uses a frontier model for dense specs, tables and long reports.
Extractors that learn from your fixes
Save a schema as an extractor for each kind of document. Correct a result once and later runs avoid the same mistake.
Batches, history and webhooks
Drop in a folder of PDFs and download one combined table. Every run is saved, and the API can run jobs in the background and call your webhook when they finish.
Scans, drawings and tables
Scanned pages, title blocks and schedules are read with AI vision, so stamped submittals and drawing sets work too.
An AI council designs your schema
Three AI models each propose a schema from a different angle: structure, domain knowledge, and accuracy. A fourth merges them into one detailed schema. No JSON Schema writing required.
A council for every subject
Save a council for invoices, another for contracts, another for listings, each with its own AI lenses, personas and models. Pick one per document.
Reads the whole document
Long pages and reports are processed in parts and merged, so every line item, table row, and section makes it into your JSON, not just the first few pages.
Terminal, Sheets and Excel
Run extractions from your dashboard without writing code, then send results to Google Sheets, Excel, CSV or SQL. PDFs up to 50 MB.
Pay only for what you use
Add credit, and each request is billed at its measured cost.
No monthly fee
Months you don’t use it cost nothing. No plan to pick, nothing to cancel.
Every feature included
The council, whole-document extraction, 50 MB PDFs and the Terminal are on from your first dollar.
See every cent
Each response includes its exact cost and your remaining balance, and your dashboard lists every charge.
Web page extraction
A typical page, one schema
~$0.004
PDF extraction, ~40 pages
~120k characters of text
~$0.03
Long document extraction
Up to ~800k characters (~250 pages)
~$0.20
High accuracy, ~40 pages
Frontier model, with sources
~$0.40 – $0.80
Scanned or drawing pages
Read with AI vision
~$0.01 per page
AI schema suggestion
3 AI lenses + a chairman model
$0.15 – $0.35
Secure checkout by Stripe · Credit doesn't expire · Runs on Cloudflare's global network
Questions
How is a request priced?
By what it actually uses: the AI tokens processed and the pages fetched. A short web page costs about $0.004; a 250-page document or an AI council suggestion costs more because it does more work. You see the exact cost of every request.
How do I know the data is right?
Turn on sources and every item carries the page and quote it came from. Jsonit checks each quote against the document and each number against its quote, and the review screen lists anything that didn’t match, so you check those instead of everything.
Is there a subscription?
No. You add credit when you need it ($10 minimum), and it doesn’t expire.
What happens when my credit runs out?
Requests are paused with a clear message until you add more. Nothing is charged to your card automatically.
What documents work?
PDFs up to 50 MB, including scanned pages and drawing sheets (read with AI vision), and public web pages. It was built for engineering and construction documents: specs, submittals, utility standards, schedules and test reports.
Stop retyping documents.
$10 of credit covers about 2,500 page extractions. Set up in two minutes.
Start with $10 of creditNot ready? Try the playground first.