Document intelligence API

Assay the page. Keep the structure.

Cupelin turns messy PDFs, scans, spreadsheets, and slides into typed, cited JSON. In-house vision models read layout the way a person would. Language models fill the schema. Heat for the dross. A vessel for what remains.

Read the docs

Line	Qty	Amount
Kiln liner, clay	12	4,800.00
Cupel lot 7B	240	1,920.00
Bead assay fee	1	85.00

{
  "vendor": "Kiln Goods LLC",
  "invoice_id": "KG-18420",
  "total": 6805.00,
  "source": { "page": 2, "box": [64, 210, 540, 318] }
}

Charge. Fire. Keep.

Method

How the cupel works

Charge.

Send a file or a batch. PDFs, scans, xlsx, images, mixed binders. Idempotency keys keep a retry from double-firing a night job.

Fire.

Vision models recover reading order, tables, handwriting, stamps, and figures. Layout stays a first-class object, with coordinates on every block.

Keep.

Extract against your schema with bounding boxes and source spans you can audit. Low-confidence fields pause for a reviewer in Studio.

Stack

Product pillars

Parse

Layout-first markdown and JSON that respects columns, headers, and nested tables.

Open Parse

Split

Break binders into the documents they actually contain.

Open Split

Extract

Typed fields, citations, and confidence you can route to a reviewer.

Open Extract

Studio

Build, eval, and version pipelines from the same engine as the API.

Open Studio

Studio

Fire small batches before you fire the warehouse.

Upload examples, lock a schema, run evals, compare versions. Studio is for the people who know the document. The canvas below is the same surface a solutions engineer uses on a new packet type.

Charge

KG-18420.pdf

binder_q3.pdf

claim_4412.tif

rates.xlsx

Page 2 of 6 · table recovered, 12 rows, nested header on columns 3 to 5.

Keep

vendor · string

total · money

line_items · array

signed_on · date

Eval v14 vs v13 94.2% exact field match · 88 rows

From the floor

Teams that live in PDFs

“We stopped babysitting OCR. Tables survive. Citations stay attached.”

Example quote, document platform lead

“Air-gapped deploy was the path legal would sign.”

Example quote, security architect

2.1B pages assayed in 2025

186 production pipelines

41.2ms median digital parse

28 people in the kiln

Metered API

Built for billed volume

Cupelin is an API you meter. Pipelines version. Evals catch regressions before a customer’s night job. Cloud, VPC, and air-gapped installs share one API surface. Assay, Foundry, and Private fire are live list prices on the pricing page.

Bring a stubborn PDF. Leave with a schema.

Rahul walks technical buyers through a live binder. Book a slot with [email protected].

Book a technical walkthrough