Question 1

How is this different from LlamaParse or Reducto?

Accepted Answer

We sit on the same parser layer (LlamaParse + Reducto + others are wrapped as facets internally), but expose a single schema-first endpoint with normalized JSON output. You write your JSON Schema once; we route + normalize. Same vendor underneath when you want it, one shape on top.

Question 2

What document types work today?

Accepted Answer

Invoices, bank statements, receipts, contracts, scanned forms, financial reports (10-K, 10-Q), insurance ACORD forms, lease agreements, scientific papers. The schema is yours — anything you can describe in JSON Schema can come back as JSON.

Question 3

What is the free tier?

Accepted Answer

Free guest extractions get a capped number of runs per browser session — enough to test against a real document. After that, sign in and mint an API key from the dashboard for unmetered usage on a prepaid credit balance.

Question 4

Can I bring my own vendor key?

Accepted Answer

Yes — once you sign in. Drop a LlamaParse or Reducto key into your account settings and we proxy your extractions to that vendor at cost, charging only the schema-normalization fee per page.

Question 5

How do you handle scanned PDFs?

Accepted Answer

Balanced and Agentic tiers run OCR before extraction. Fast tier skips OCR and is intended for born-digital PDFs where text is already embedded — use it on invoices and contracts, not photos of paper.

Question 6

What about citations and confidence scores?

Accepted Answer

Every extracted field can come back with a page number and bounding box pointing to the source span. Add `confidence` to your schema property and we return per-field confidence scores.

Question 7

Is the playground bound to my account?

Accepted Answer

Anonymous runs use the per-session free tier and are not persisted. Signed-in runs are saved to your account and bill against your prepaid credits.

Question 8

How do I deploy this in production?

Accepted Answer

Copy the cURL/TS/Python sample, add your API key as a Bearer header, point it at the same endpoint with your own file_id from POST /v1/files. Documented in the SDK quickstart.