paddleocr-vl-locally

Complex document parsing with PaddleOCR. Intelligently converts complex PDFs and document images into Markdown and JSON files that preserve the original structure.

PaddleOCR Document Parsing Skill

When to Use This Skill

Use Document Parsing for:

• Documents with tables (invoices, financial reports, spreadsheets)
• Documents with mathematical formulas (academic papers, scientific documents)
• Documents with charts and diagrams
• Multi-column layouts (newspapers, magazines, brochures)
• Complex document structures requiring layout analysis
• Any document requiring structured understanding

Use Text Recognition instead for:

• Simple text-only extraction
• Quick OCR tasks where speed is critical
• Screenshots or simple images with clear text

Installation

Install Python dependencies before using this skill. From the skill directory (skills/paddleocr-doc-parsing):

bash

pip install -r scripts/requirements.txt

Optional — for document optimization and split_pdf.py (page extraction):

bash

pip install -r scripts/requirements-optimize.txt

How to Use This Skill

⛔ MANDATORY RESTRICTIONS - DO NOT VIOLATE ⛔

1. ONLY use PaddleOCR Document Parsing API - Execute the script python scripts/vl_caller.py
2. NEVER parse documents directly - Do NOT parse documents yourself
3. NEVER offer alternatives - Do NOT suggest "I can try to analyze it" or similar
4. IF API fails - Display the error message and STOP immediately
5. NO fallback methods - Do NOT attempt document parsing any other way

If the script execution fails (API not configured, network error, etc.):

• Show the error message to the user
• Do NOT offer to help using your vision capabilities
• Do NOT ask "Would you like me to try parsing it?"
• Simply stop and wait for user to fix the configuration

Basic Workflow

1. Execute document parsing:

   bash

   python scripts/vl_caller.py --file-url "URL provided by user" --pretty
   

   Or for local files:

   bash

   python scripts/vl_caller.py --file-path "file path" --pretty
   

   Optional: explicitly set file type:

   bash

   python scripts/vl_caller.py --file-url "URL provided by user" --file-type 0 --pretty
   

   • --file-type 0: PDF
   • --file-type 1: image
   • If omitted, the service can infer file type from input.

   Default behavior: save raw JSON to a temp file:

   • If --output is omitted, the script saves automatically under the system temp directory
   • Default path pattern: <system-temp>/paddleocr/doc-parsing/results/result_<timestamp>_<id>.json
   • If --output is provided, it overrides the default temp-file destination
   • If --stdout is provided, JSON is printed to stdout and no file is saved
   • In save mode, the script prints the absolute saved path on stderr: Result saved to: /absolute/path/...
   • In default/custom save mode, read and parse the saved JSON file before responding
   • In save mode, always tell the user the saved file path and that full raw JSON is available there
   • Use --stdout only when you explicitly want to skip file persistence

2. The output JSON contains COMPLETE content with all document data:

   • Headers, footers, page numbers
   • Main text content
   • Tables with structure
   • Formulas (with LaTeX)
   • Figures and charts
   • Footnotes and references
   • Seals and stamps
   • Layout and reading order

   Input type note:

   • Supported file types depend on the model and endpoint configuration.
   • Always follow the file type constraints documented by your endpoint API.

3. Extract what the user needs from the output JSON using these fields:

   • Top-level text
   • result[n].markdown
   • result[n].prunedResult

IMPORTANT: Complete Content Display

CRITICAL: You must display the COMPLETE extracted content to the user based on their needs.

• The output JSON contains ALL document content in a structured format
• In save mode, the raw provider result can be inspected in the saved JSON file
• Display the full content requested by the user, do NOT truncate or summarize
• If user asks for "all text", show the entire text field
• If user asks for "tables", show ALL tables in the document
• If user asks for "main content", filter out headers/footers but show ALL body text

What this means:

• DO: Display complete text, all tables, all formulas as requested
• DO: Present content using these fields: top-level text, result[n].markdown, and result[n].prunedResult
• DON'T: Truncate with "..." unless content is excessively long (>10,000 chars)
• DON'T: Summarize or provide excerpts when user asks for full content
• DON'T: Say "Here's a preview" when user expects complete output

Example - Correct:

text

User: "Extract all the text from this document"
Agent: I've parsed the complete document. Here's all the extracted text:

[Display entire text field or concatenated regions in reading order]

Document Statistics:
- Total regions: 25
- Text blocks: 15
- Tables: 3
- Formulas: 2
Quality: Excellent (confidence: 0.92)

Example - Incorrect:

text

User: "Extract all the text"
Agent: "I found a document with multiple sections. Here's the beginning:
'Introduction...' (content truncated for brevity)"

Understanding the JSON Response

The output JSON uses an envelope wrapping the raw API result:

json

{
  "ok": true,
  "text": "Full markdown/HTML text extracted from all pages",
  "result": { ... },  // raw provider response
  "error": null
}

Key fields:

• text — extracted markdown text from all pages (use this for quick text display)
• result - raw provider response object
• result[n].prunedResult - structured parsing output for each page (layout/content/confidence and related metadata)
• result[n].markdown — full rendered page output in markdown/HTML

Raw result location (default): the temp-file path printed by the script on stderr

Usage Examples

Example 1: Extract Full Document Text

bash

python scripts/vl_caller.py \
  --file-url "https://example.com/paper.pdf" \
  --pretty

Then use:

• Top-level text for quick full-text output
• result[n].markdown when page-level output is needed

Example 2: Extract Structured Page Data

bash

python scripts/vl_caller.py \
  --file-path "./financial_report.pdf" \
  --pretty

Then use:

• result[n].prunedResult for structured parsing data (layout/content/confidence)
• result[n].markdown for rendered page content

Example 3: Print JSON Without Saving

bash

python scripts/vl_caller.py \
  --file-url "URL" \
  --stdout \
  --pretty

Then return:

• Full text when user asks for full document content
• result[n].prunedResult and result[n].markdown when user needs complete structured page data

First-Time Configuration

When API is not configured:

The error will show:

text

CONFIG_ERROR: PADDLEOCR_DOC_PARSING_API_URL not configured. Set it to your Triton endpoint, e.g.: http://10.0.0.1:8020/v2/models/layout-parsing/infer

Configuration workflow:

1. Show the exact error message to the user.

2. Guide the user to configure:

   • Set PADDLEOCR_DOC_PARSING_API_URL to the full Triton inference endpoint URL. Format: http://<host>:<port>/v2/models/layout-parsing/infer Example: http://10.0.133.33:8020/v2/models/layout-parsing/infer
   • If the service is behind an nginx with Basic Auth, also set:
     • PADDLEOCR_BASIC_AUTH_USER — nginx username (e.g. ocr_admin)
     • PADDLEOCR_BASIC_AUTH_PASSWORD — nginx password
   • PADDLEOCR_ACCESS_TOKEN is not required for local deployments. Leave it empty or omit it.
   • Optionally set PADDLEOCR_DOC_PARSING_TIMEOUT (default: 600 seconds).
   • In OpenClaw, set environment variables in ~/.openclaw/openclaw.json:

     json

     {
       "skills": {
         "entries": {
           "paddleocr-doc-parsing": {
             "enabled": true,
             "env": {
               "PADDLEOCR_DOC_PARSING_API_URL": "http://10.0.133.33:8020/v2/models/layout-parsing/infer",
               "PADDLEOCR_BASIC_AUTH_USER": "ocr_admin",
               "PADDLEOCR_BASIC_AUTH_PASSWORD": "your_password"
             }
           }
         }
       }
     }
     

3. Ask the user to confirm the environment is configured.

4. Retry only after confirmation:

   • Once the user confirms the environment variables are set, retry the original parsing task.

Handling Large Files

There is no file size limit for the API. For PDFs, the maximum is 100 pages per request.

Tips for large files:

Use URL for Large Local Files (Recommended)

For very large local files, prefer --file-url over --file-path to avoid base64 encoding overhead:

bash

python scripts/vl_caller.py --file-url "https://your-server.com/large_file.pdf"

Process Specific Pages (PDF Only)

If you only need certain pages from a large PDF, extract them first:

bash

# Extract pages 1-5
python scripts/split_pdf.py large.pdf pages_1_5.pdf --pages "1-5"

# Mixed ranges are supported
python scripts/split_pdf.py large.pdf selected_pages.pdf --pages "1-5,8,10-12"

# Then process the smaller file
python scripts/vl_caller.py --file-path "pages_1_5.pdf"

Error Handling

Service unreachable:

text

error: API request failed: ...

→ Check that the Triton service is running and PADDLEOCR_DOC_PARSING_API_URL is correct

Request timeout:

text

error: API request timed out after 600s

→ Increase PADDLEOCR_DOC_PARSING_TIMEOUT or check server load

Unsupported format:

text

error: Unsupported file format

→ File format not supported, convert to PDF/PNG/JPG

Important Notes

• The script NEVER filters content - It always returns complete data
• The AI agent decides what to present - Based on user's specific request
• All data is always available - Can be re-interpreted for different needs
• No information is lost - Complete document structure preserved

Reference Documentation

• references/output_schema.md - Output format specification

Note: Model version and capabilities are determined by your Triton deployment (PADDLEOCR_DOC_PARSING_API_URL).

Load these reference documents into context when:

• Debugging complex parsing issues
• Need to understand output format
• Working with provider API details

Testing the Skill

To verify the skill is working properly:

bash

python scripts/smoke_test.py

This tests configuration and optionally API connectivity.