ocr-local

What This Skill Does

The ocr-local skill provides a robust, private, and efficient text recognition engine for the OpenClaw platform. By utilizing Tesseract.js, it performs Optical Character Recognition (OCR) entirely on your local machine. This means your images are never uploaded to a cloud server, ensuring complete data privacy and security. The tool supports high-accuracy detection of Simplified Chinese, Traditional Chinese, and English, making it ideal for extracting data from screenshots, PDFs, or photos of documents without needing expensive or rate-limited API keys.

Installation

To integrate this capability into your workflow, use the following OpenClaw command in your terminal:

clawhub install openclaw/skills/skills/15914355527/ocr-local-v2

Once installed, the skill resides within your local scripts directory. Note that the first time you execute an OCR task, the engine will automatically download the necessary language training data (approximately 20MB per language). These files are cached locally for future use, ensuring that subsequent requests remain extremely fast.

Use Cases

• Screenshot Data Extraction: Convert text found in application UI, dashboards, or error messages into searchable/copyable text.
• Document Digitization: Transform images of receipts, invoices, or manual notes into digital text for archiving or database entry.
• Multilingual Support: Extract mixed-language data from documents containing both English and Chinese characters.
• Automation Pipelines: Chain this skill with other OpenClaw tools to perform text-based analysis on image assets automatically.

Example Prompts

1. "OpenClaw, extract the text from the screenshot named invoice_001.png and save it as a text file for my records."
2. "Use the ocr-local skill on document.jpg and translate the content into English if it contains Traditional Chinese."
3. "Scan the dashboard image and give me the total value figures found in the text using the JSON output mode."

Tips & Limitations

For optimal results, ensure your input images are of high resolution and contrast. While the model is highly capable, it may struggle with low-resolution photos, complex background patterns, or heavily stylized handwriting. Because it runs locally, processing time is dependent on your machine's CPU power; large images may take a few seconds longer to process. Always ensure your image files are accessible to the script's path when calling it via the command line.