FROM UPLOAD TO USABLE DATA
01
Define what good looks like
Write validation criteria in plain language so irrelevant, incomplete, or unreadable files are caught early.
02
Describe the data you need
Choose text, number, date, or list fields. Add hints when wording, format, or location could be ambiguous.
03
Test until you trust it
Upload a PDF, JPG, or PNG. Review validation, image quality, confidence, and every extracted value in one place.
04
Put the result to work
Map values into case data, trigger extraction on upload, or send and retrieve jobs through the Penbox API.
VALIDATE
Catch the wrong document first
Check each upload against criteria written in plain language and return a clear validation verdict.
STRUCTURE
Return data in the right shape
Extract text, numbers, dates, and repeatable lists instead of loose blocks of text that need another pass.
GUIDE
Resolve ambiguity with field hints
Tell the model which date, format, region, or value matters when the same document contains several candidates.
TRUST
See confidence and image quality
Review field-level confidence, blur detection, document type, and a human-readable validation explanation.
REUSE
Standardize every document type
Create reusable templates for IDs, invoices, bank details, certificates, contracts, or any recurring document class.
REASON
Understand the whole document
Use Advanced when handwriting, checkboxes, sketches, or multi-page context must be read together to infer the answer.
DOCUMENT INTELLIGENCE 3
Fast
DEFAULT
3 credits per page
For printed and structured documents where values appear in predictable places: IDs, bank details, invoices, policy schedules, and certificates.
30% faster
than the previous model
25% more accurate
than the previous model
Advanced
ON REQUEST
6 credits per page
For documents Fast has to guess at: handwriting, checkbox grids, sketches, inference, and context that spans several pages.
50% more accurate
than the previous model
Only 10% slower
than the new Fast model
FROM EXTRACTION TO PROGRESS
Operators keep the document and the structured result together, inside the same case timeline and audit trail.
PENBOX AGENT
Build and refine templates by describing what you need.
Add a list field for line items with description, quantity, unit price, and total.
Add, rename, remove, and reorder fields. Set types, define list sub-fields, add hints, and generate stable keys without leaving the template.
PENBOX API
Run extraction from your own systems.
POST /document_intelligence
GET /document_intelligence/{id}
Send existing attachments or new files, reference a template or provide fields directly, add a language hint, and retrieve typed values, confidence, validation, document type, and page usage.
The model changes. Your extraction schema does not.