A high-precision document parsing pipeline that extracts layout elements and performs deep multimodal analysis on tables and charts.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install pdf-ocr-layout
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install pdf-ocr-layout using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is GLM-OCR Multimodal Document Deep Analysis?
The GLM-OCR Multimodal Deep Analysis skill is a sophisticated document parsing pipeline designed for high-precision extraction and interpretation of complex layouts. By leveraging Zhipu's GLM-OCR for physical layout analysis, GLM-4.7 for logical table interpretation, and GLM-4.6V for multimodal visual analysis, this tool transforms raw PDFs and images into structured, meaningful data.
Unlike traditional OCR tools that merely digitize text, this Openclaw Skills asset understands the relationship between visual elements and their context. It identifies tables, crops illustrations as independent files, and provides deep semantic reasoning to explain the content of your documents in a way that is immediately actionable for developers and analysts.
GLM-OCR Multimodal Document Deep Analysis Use Cases
- Extracting complex financial tables from PDF reports and converting them to Markdown with high logical precision.
- Automatically identifying and cropping architectural diagrams or charts from technical manuals for independent storage.
- Generating deep semantic descriptions of visual charts by combining image data with surrounding textual context.
- Building automated document processing workflows that require reasoning beyond simple text recognition.
How GLM-OCR Multimodal Document Deep Analysis Works
- The input file is processed through the Extraction Stage where GLM-OCR performs a physical layout analysis of the document.
- Layout elements such as tables and images are identified, and their bounding box (Bbox) coordinates are mapped.
- Table HTML is extracted and cleaned into Markdown format, while charts and illustrations are automatically cropped into independent image files.
- In the Understanding Stage, text-based elements like tables are passed to GLM-4.7 for logical interpretation within the document's full context.
- Visual elements like charts are processed by GLM-4.6V, providing a multimodal analysis that explains the image's meaning in relation to the text.
- The final output is a structured JSON report containing the original content, layout metadata, and the deep understanding generated by the AI models.
GLM-OCR Multimodal Document Deep Analysis Setup
Ensure you have Python 3.8 or higher installed. Start by configuring your Zhipu AI environment variable:
export ZHIPU_API_KEY='your_api_key_here'
Install the required Python dependencies:
pip install zhipuai pillow beautifulsoup4
Run the pipeline using the following command to process your files:
python scripts/glm_ocr_pipeline.py --file_path "/path/to/your/document.pdf" --output_dir "/path/to/output"
GLM-OCR Multimodal Document Deep Analysis Data Schema & Taxonomy
The skill generates an output directory containing cropped images and returns a JSON array of objects. Each object represents a layout element with the following structure:
| Key | Type | Description |
|---|---|---|
type |
String | The element category: table or image. |
bbox |
Array | Coordinates [x1, y1, x2, y2] of the element on the page. |
content_info |
String | The Markdown table content or the file path to the cropped image. |
deep_understanding |
String | AI-generated semantic analysis based on GLM-4.7 or GLM-4.6V. |
GLM-OCR Multimodal Document Deep Analysis Advanced Features
- Context-aware reasoning that analyzes tables and charts relative to the entire document text rather than in isolation.
- Automatic image cropping based on Bbox coordinates for high-resolution extraction of visual assets.
- Multimodal model routing that dynamically switches between GLM-4.7 and GLM-4.6V depending on content type.
- Full page reading order preservation via intermediate JSON generation during the extraction stage.
SKILL.md
Loading
Related Openclaw Skills
Feishu Leave Request
A browser automation skill that facilitates the end-to-end submission of leave requests within the Feishu (Lark) corporate ecosystem.
Trading Analyzer
A unified market analysis engine for cryptocurrency and stocks powered by multi-source financial data and technical indicators.
byt-workflow
A sophisticated automation pipeline that downloads YouTube audio and captures real-time AI translations using the Doubao interface.
XMind Generator & Reader
A chat-first skill to generate, read, and summarize XMind mind maps directly from your AI agent using the xmind-generator-mcp server.
MinerU PDF Processor
A local document parsing tool that converts PDFs into machine-readable Markdown and JSON while preserving complex structures like formulas and tables.
Turkiye AFAD Deprem Skill
A specialized tool for retrieving and filtering Turkish earthquake data from official AFAD sources.
METADATA
Featured*
