Multi-Engine Document Parsing

The right parsing engine for every document

Vision AI for visual layouts, Text AI for plain text, templates for fixed forms. All three engines run in the same mailbox.

What's included

Vision AI extraction

Vision models read pages as images rather than as text. The AI sees the document the way a human reader would, with full layout and visual context.

Best for rich PDFs, scans, and forms with complex structure
Captures handwriting, checkboxes, stamps, and layout cues
Set up with plain-English instructions, no template required

Text AI extraction

Documents are first converted to plain text, using OCR when the source has no native text layer. The AI then parses the extracted text alone, ignoring layout and images.

Best for emails, plain PDFs, and other text-first documents
Useful when visual layout adds no information
Set up with plain-English instructions, no template required

Template-based extraction

Add as many templates per mailbox as you need. Parseur auto-picks the best match per document, producing the same output every time, no AI involved.

Best for standardized forms and machine-generated emails
The most reliable extraction method when layouts never change
Set up with a visual template editor, one per document layout

Table and line item extraction

Each row in a table becomes its own data record, not a single merged field. Works across all three parsing engines. For native spreadsheets, table parsing is automatic.

Handles variable row counts from one document to the next
Supports tables that span multiple pages
AI engines support parsing complex, multi-line rows into separate fields

OCR for scanned documents and images

Optical Character Recognition reads text from scans, phone photos, and image-only PDFs. Feeds the Text AI and template engines when there's no native text layer.

Works on scans, phone photos, and image-only PDFs
Multilingual OCR across 200+ languages, including handwriting
Template engine uses zonal and dynamic OCR for fixed or shifting layouts

Document pre-processing

Accurate extraction starts with cleaning up and repairing incoming documents. Parseur's pre-processing has been refined over 100M+ documents and a decade of real-world edge cases.

Deskews tilted scans and re-runs OCR on garbled text
Repairs corrupted PDFs, broken email encoding, and malformed HTML
Detects country-specific date and number formats automatically

How Document Parsing works

What just happened

Document Intake

Documents were uploaded or arrived automatically from email, API, or connected storage.

Learn more

Pre-process

Each document goes through a cleanup pass first. Parseur fixes page orientation, straightens tilted scans, and repairs garbled or misordered content as needed.

9° skewed

Ready

OCR

For scans, phone photos, and image-only PDFs, Parseur runs OCR to extract the text. Documents that already carry a native text layer skip this step.

INVOICE #Q2-8821

Acme Corp

April 15, 2026

Due May 15

Sender

Acme Corp

acme.com

Total

$163.00

USD

OCR scanning

Pick the engine

Parseur picks the right engine for each document automatically. Template-based parsing takes priority when a matching template exists, otherwise Vision AI handles image-rich pages and Text AI handles plain text content.

Template

AI Vision

AI Text

Extract

The selected parsing engine pulls structured fields out of the document, mapped to the schema you defined in your mailbox. From here, every field flows into normalization for formatting and validation.

INVOICE #Q2-8821 Invoice no

Acme Corp

Customer

May 7, 2026

Date

Due in 30 days

Sender

Acme Corp

acme.com

Bill to

Globex Inc

Springfield

Items Item Qty Price Consulting 2 $50 Equipment 1 $25 Setup fee 3 $73

Subtotal $148.00

Tax $15.00

Total $163.00 Total

Extracting

What happens next

Data Normalization and Validation

Extracted fields are validated, formatted, and shaped for downstream workflows.

Learn more

Back to all features

Document parsing on autopilot.

Upload a sample, name the fields you need, and watch Vision AI, Text AI, or templates do the work.

Free plan included, no credit card needed

Process your first document in under 2 minutes

Cancel anytime, no commitment

Frequently Asked Questions

Common questions about Parseur's parsing engines, from Vision AI and OCR to templates, table extraction, and multi-language support.

Document parsing is the process of pulling structured fields out of unstructured documents like PDFs, scans, or emails, so the data can be used in spreadsheets, databases, and connected tools without manual re-keying. Parseur runs three parsing engines, Vision AI, Text AI, and templates, and picks the right one per document automatically.

Vision AI reads pages as images and uses full layout context, including handwriting, checkboxes, stamps, and visual cues. It is best for rich PDFs, scans, and forms with complex structure. Text AI works on the plain text of a document, ignoring layout, and is best for emails, plain PDFs, and other text-first content.

Yes. A mailbox can hold as many templates as you need, one per document layout. When a new document arrives, Parseur automatically picks the best matching template, so a single mailbox can handle many fixed layouts side by side. If no template matches, Vision AI or Text AI takes over so the document is still parsed.

Yes. Scans, phone photos, and image-only PDFs are handled by built-in OCR, and Vision AI captures handwriting, checkboxes, stamps, and other visual elements that text-only tools miss.

Yes. Each row of a table becomes its own data record rather than a merged blob of text. Table extraction works across all three parsing engines, supports variable row counts, and handles tables that span multiple pages. Native spreadsheets are parsed as tables automatically.

Accuracy depends on the engine and the document. Templates produce identical output every time on fixed layouts. Vision AI handles complex visual structure, Text AI handles plain text. Pre-processing repairs tilted scans, garbled text, broken encoding, and corrupted PDFs before extraction, and downstream validation catches issues before data leaves Parseur.

You upload a sample document and Parseur auto-identifies the fields it thinks you want extracted. From there, you can refine the field list and write plain-English instructions per field. The AI uses those instructions to extract the right values from new incoming documents, even when layouts vary. No model training or custom code is needed.

No. Both Vision AI and Text AI work with plain-English instructions and require no template. Templates are still available for fixed layouts where you want guaranteed identical output every time, like machine-generated forms.

Yes. OCR runs automatically on scans, phone photos, and image-only PDFs to extract a text layer for the parsing engines. Documents that already carry a native text layer skip the OCR step.

OCR works across 200+ languages, including handwriting. The AI engines understand documents in any major language, and country-specific date and number formats are detected automatically from document context.

Yes. Vision AI and Text AI adapt to layout variation without per-vendor templates, so a single mailbox can process invoices or receipts from many different senders with their own formats.

Sign up, create a mailbox, and drop a sample PDF in. On first upload, Parseur identifies the fields it thinks you want extracted. You can adjust the list of fields and the plain-English instructions at any time after that. The parsing engine is picked automatically per document, and parsed data can be sent to Google Sheets, your CRM, a database, or any custom endpoint without writing code.