Parseur offers multiple AI parsing engines optimized for different document types and extraction scenarios.
You can choose the AI engine for each mailbox in Mailbox Settings > General > AI engine:
Available AI Engines
Engine | Best For | Key Strengths | Limitations |
AI Text Engine v3 | Advanced text extraction and evolving AI features | Improved reasoning, long-document understanding | Cannot process visual data like handwriting, checkboxes, etc. |
AI Vision Engine v3 (Default) | Complex PDFs, scans, images, and visually rich documents | Understands layout, visual structure, handwriting, checkboxes, stamps, and multi-column documents | Can be unnecessary for simple text documents |
AI Engine v4 (Experimental) | All documents | Combines both Vision and Text Engines. Better accuracy across the board, faster processing of larger documents | Current best option |
Disabled | Template-only workflows | Ensures only templates are used | No AI fallback for unmatched documents |
AI Text Engine v3
Recommended for users testing the latest AI parsing improvements.
AI Text engine v3 is Parseur’s newer experimental text AI engine. It includes improvements in reasoning, instruction following, and long-document understanding.
Best use cases
Complex text extraction scenarios
Long documents
Documents requiring detailed instructions
Testing newer AI capabilities
Advantages
Better understanding of long documents
Improved instruction following
More advanced contextual reasoning
Improved extraction flexibility
Limitations
Experimental engine
Some edge cases may behave differently between updates
Not always ideal for highly stable production workflows
Recommended when
Use AI Text engine v3 when:
You want to test the latest parsing improvements
Your extraction logic relies heavily on instructions
Documents contain large amounts of text or complex context
AI Vision Engine v3 (Default)
Recommended for most standard use cases as well as visually complex documents and scanned files.
AI Vision engine v3 uses vision-language models (VLMs) to analyze documents visually, similarly to how a human reads a page. This is our most stable engine.
Unlike text-only engines, Vision AI understands:
Layout structure
Visual positioning
Handwriting
Checkboxes
Highlighted fields
Tables with inconsistent formatting
Stamps and signatures
Best use cases
Scanned PDFs
Photos of documents
Multi-column layouts
Forms with checkboxes
Handwritten content
Visually rich documents
Low-quality scans
Documents where layout matters
Advantages
Understands both text and visual structure
Better extraction from difficult layouts
Fastest processing speed
Handles layout interpretation together
Stronger performance on image-based documents
Better handling of messy or inconsistent formatting
Limitations
Can be unnecessary for simple text documents
Recommended when
Use AI Vision engine v3 when:
Documents are scanned or image-based
Visual structure matters
Text extraction alone is insufficient
You need to capture checkboxes, signatures, or handwriting
Documents contain complex layouts
AI Engine v4 Experimental
Recommended for all files.
A blend of both Text and Vision engines, this method of extraction should work best for all documents. Choose this engine if you are getting unexpected results with the other two.
We are currently collecting feedback on this version, please let us know your thoughts at [email protected]
Disabled Mode
When AI is disabled, Parseur will only use template-based parsing.
If no matching template is found, the document will fail processing.
Recommended when
Disable AI when:
You require deterministic extraction
All incoming documents follow known layouts
You only want template-based parsing
Compliance or workflow requirements prohibit AI parsing
Which AI Engine Should You Choose?
Choose AI Text engine v3 if:
You want newer AI capabilities
You process long or context-heavy documents
You rely heavily on AI instructions
You are comfortable testing experimental improvements
Choose AI Vision engine v3 if:
You want the most stable experience
Processing speed is paramount
Your documents are scanned or image-based
Layout and visual context matter
You need handwriting or checkbox extraction
Documents contain complex formatting
Disable AI if:
You exclusively use templates
Your layouts are fixed and predictable
You require deterministic extraction behavior
Important Notes
Matching templates always take priority over AI engines.
If no template matches, Parseur will use the selected AI engine.
For best results, use clear field names and concise instructions.
AI engines can work alongside template-based workflows in the same mailbox.