Skip to main content
The Parse feature transforms complex documents into machine-readable data, optimized for LLMs.
Parse Viewer
It intelligently identifies document elements, processes them based on their type, and outputs clean HTML & Markdown content ready for AI applications and downstream workflow automation.

Key Features

  • Perfect Markdown & HTML: LLM-ready content (Markdown, HTML, tables, etc).
  • Reading order intact: Maintains the natural reading flow for complex layouts.
  • Granular bounding boxes: Pinpoints element coordinates with precision for easy citations.
  • Native Spreadsheet handling: 100% reconstruction with formulas, styling, and cell values preserved; precise ranges; cleans tables and converts charts to structured data.
  • Post-processing: Token-aware chunking, cropped images, and more.

Example: Parse and access chunk content

Here’s how you can parse a document and access its chunks using our SDKs.

Our default configuration is optimized through extensive testing and provides excellent results for most documents. You can customize parse if you have specific requirements.
For a comprehensive breakdown of every available configuration, please refer to our API Reference. Here is an overview of our configuration options:
  • Pipeline (pipeline): Choose the provider (Azure or Chunkr) for layout analysis and OCR models.
  • Layout Analysis & OCR:
    • Segmentation Strategy (segmentation_strategy): Choose between LayoutAnalysis (default) or a full-page VLM approach for parsing.
    • OCR Strategy (ocr_strategy): Use Auto to selectively apply OCR or All to force it on every page.
  • Segment-level Customization (segment_processing): Control processing for each document element (e.g., Text, Table, Picture):
    • Processing Strategy (strategy): For each segment, set the strategy to generate HTML/Markdown. Auto (simple OCR + logic), LLM (VLM generation), or Ignore (remove from output).
    • Format Control (format): Control the output format (Markdown or HTML) for segment content.
    • Extended Context (extended_context): Provide the full page image as additional context for VLM processing of a segment. Useful for cases like distant legends for tables and pictures.
    • Cropped Images (crop_image): Control if a cropped image of the segment is included.
  • Chunking (chunk_processing): Configure chunking strategy, sizes, and token-counting model.
  • Error Handling (error_handling): Set to Fail (default) to stop on any error, or Continue to process despite non-critical errors.