Limitations
A score is not a safety certificate
Detection depends on format and parser support. Low or zero risk does not prove a file is private. Visible names, screenshots, faces and private text remain. Nonstandard metadata and covert data channels may be missed.
Images
JPEG, PNG, static WebP and supported single-photo HEIC are accepted within 25 MB and 24 megapixels. The screenshot editor uses a separate 8 megapixel/50-mask budget. Image cleaning re-encodes decoded pixels at 95% JPEG quality or lossless PNG. Colors, quality, orientation handling and auxiliary content can vary by browser. PNG compressed text chunks are detected, not decompressed. Animated PNGs become a single decoded frame.
PDFs
PDFs are limited to 50 MB and 300 pages. Encrypted files are rejected. Metadata cleaning preserves annotations, forms, attachments and scripts by default. Explicit options remove supported annotations, embedded files and actions, with payload pruning and page-stream/form-value integrity checks. Redaction marks block annotation removal; XFA and other uninspected data can remain. Invisible text detection is a limited redaction signal; covered text, clipping, layers and image redactions require manual verification.
DOCX
DOCX inspection covers properties, comments, revision elements, hidden-format markers, embedded parts, media inventories and relationships. Embedded media metadata and full visible content are not analyzed. Cleaning removes core/custom properties and selected extended fields. Comments and revisions remain by default; explicit review-copy choices can remove supported review data or alter text. Complex revisions are refused for acceptance/rejection. External links and embeddings remain except for exclusively owned removed comment resources. Structural reopening cannot guarantee appearance in all Office versions. ZIP64, non-UTF-8 XML and unusual packages may be rejected.
Browser limits
Current browsers with Web Workers are required; photo cleaning additionally requires OffscreenCanvas and createImageBitmap. Processing stops after 10 seconds for ordinary image/DOCX checks or 30 seconds for PDF/HEIC work and screenshot exports. Batches are limited to 10 files, 100 MB total selection and 100 MB retained clean output, and run sequentially. The browser may still run out of memory on unusual files.