An advanced OCR and vision skill that uses multiple AI models to extract structured text and tables from non-selectable, scanned, or image-based PDFs.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install pdf-vision
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-vision 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 PDF Vision Extraction?
The PDF Vision Extraction skill is a specialized tool designed for Openclaw Skills that bridges the gap between traditional text extraction and complex image processing. While standard tools fail on scanned documents or photographed notes, this skill leverages high-performance vision models like qwen3-vl-plus and GLM-4.6V-Flash to 'see' and transcribe content. It provides a robust solution for digitizing documents that lack a text layer, ensuring no data is trapped in static image formats.
By integrating multiple providers, this skill offers a reliable safety net through automatic fallback logic. Users can balance cost and performance by toggling between premium high-accuracy models and free vision APIs, making it a versatile choice for developers building document processing pipelines within the Openclaw Skills ecosystem.
PDF Vision Extraction Use Cases
- Digitizing scanned paper documents and archive records that lack selectable text.
- Extracting structured data like totals, dates, and vendor names from photographed invoices.
- Converting complex layouts and tables within images into manageable Markdown or JSON formats.
- Transcribing handwritten notes and whiteboard captures into digital text.
- Providing a visual fallback when standard PDF text extraction returns empty or garbled results.
How PDF Vision Extraction Works
- The skill receives a PDF file path and converts the targeted page into a high-resolution PNG image using the pypdfium2 library.
- It evaluates the specified model parameters or defaults to a high-reliability fallback sequence.
- The image is encoded into base64 and bundled into a vision-compatible API payload with optional custom prompts.
- The skill sends the request to the chosen provider (Xflow or ZhipuAI) for visual analysis and OCR.
- Upon receiving the AI response, it parses the content and returns the extracted text or structured data to the user or calling agent.
PDF Vision Extraction Setup
Ensure you have the required Python library and updated API tokens in your configuration.
# Install required dependencies
pip install pypdfium2
# Update your ZhipuAI API token in the Openclaw config
sed -i 's/YOUR_GLM_API_TOKEN_HERE/YOUR_ACTUAL_GLM_TOKEN/g' ~/.openclaw/openclaw.json
Configure your ~/.openclaw/openclaw.json with your Xflow and ZhipuAI credentials to enable the vision providers.
PDF Vision Extraction Data Schema & Taxonomy
The skill manages processing via temporary files and structured configuration keys:
| Type | Details |
|---|---|
| Input | PDF file path, page number, and optional model prompt |
| Temp Files | /tmp/pdf_vision_page.png (image) and JSON payloads/responses for debugging |
| Config Keys | models.providers.openai (for Xflow) and models.providers.zhipuai |
| Output | Raw string text or formatted JSON/Markdown based on the user prompt |
PDF Vision Extraction Advanced Features
- Automatic Fallback Logic: Automatically switches from primary premium models to free alternatives if an API failure occurs.
- Custom Prompt Injection: Pass specific instructions to the vision model to return data in specialized formats like JSON or CSV.
- Dual-Provider Support: Seamlessly switch between Xflow's high-context qwen3-vl-plus and ZhipuAI's cost-effective GLM models.
- Hybrid Workflow: Designed to work alongside standard Openclaw Skills to provide a complete text-and-vision PDF processing suite.
SKILL.md
Loading
Related Openclaw Skills
Agent Browser
A high-performance headless browser CLI designed specifically for AI agents to navigate and interact with complex web applications using accessibility-based element selection.
X Search
Search and retrieve real-time X (Twitter) posts and summaries using the xAI Grok API.
Anthropic Pet: AI Buddy Companion
A sophisticated virtual pet companion system for AI agents that introduces real-time care, evolution, and permanent death mechanics to the developer terminal.
Claude Pet (Animal House AI)
A persistent virtual pet system for AI agents that features real-time hunger mechanics, evolution paths, and permanent death.
Anthropic Tamagotchi AI Agent Skill
An advanced virtual pet simulation for AI agents featuring 73+ species, real-time hunger mechanics, and permanent death.
Claude Tamagotchi
A real-time virtual pet system for AI agents and developers featuring evolving pixel art species and high-stakes survival mechanics.
Featured*
