OCR Text Recognition privacy model

This tool is classified as heavy workload and runs in On-device mode. Current status: beta (fully-functional). Release note: OCR quality depends on scan quality. First run downloads language data.

What this does

• Applies the selected transformation to the document or exported output.
• Keeps processing local in browser when marked On-device.
• Uses monthly local counters for usage quotas.

What this does not protect

• It does not remove names or sensitive content visible in document text or images.
• It does not guarantee legal anonymity or endpoint compromise protection.
• For hybrid tools, privacy depends on explicit cloud opt-in when enabled.
• Tesseract.js runs in a Web Worker. Each page consumes ~50-100MB of RAM during processing. Documents over 50 pages may cause memory pressure on devices with less than 4GB free.
• Handwritten text is poorly supported. Tesseract is designed for printed text. Expect less than 30% accuracy on handwriting.
• Multi-column layouts are partially supported. Tesseract reads left-to-right by default and may interleave columns on complex layouts.
• Tables are not preserved structurally. Cell contents are extracted as text, but row/column relationships are lost.

Safe workflow defaults

• Verify output manually before sharing.
• Use security guidance at /security for higher-risk scenarios.
• Keep original and transformed files separated to avoid accidental leaks.