Capabilities and limitations

SandboxPDF processes supported documents on your device. Different operations preserve different parts of a PDF, so choose the tool and inspect its output with the intended use in mind.

Pages, layout and document structure

Merge, split, extract, organise and rotate tools work with PDF pages. N-up and booklet tools arrange embedded pages onto new sheets. Page content can remain sharp, but a rebuilt document should not be assumed to preserve every original bookmark, attachment, interactive feature or digital signature. Cropping hides content outside a page boundary; it is not permanent redaction.

Redaction and metadata

Permanent redaction rebuilds pages from rendered images after applying the marked regions. Original text objects, attachments and hidden structures are discarded. This also removes searchability and interactivity. Text snapping uses extractable text runs; on scans, mark the complete sensitive region manually.

Metadata cleaning removes supported PDF properties and structures. The text-preserving mode does not promise to remove EXIF inside embedded images or sensitive content visible on pages. The visual sanitising mode discards original objects by rebuilding rendered pages. Inspect the exported copy before sharing it.

Office conversions

Word, Excel and PowerPoint imports use a local LibreOffice engine, including supported tables, images, fonts and print layouts. Missing fonts and complex layouts can need correction. PDF-to-Word and PDF-to-PowerPoint reconstruct positioned editable text over page graphics. PDF-to-Excel infers columns from text positions; PDFs do not contain the original spreadsheet formulas. Image-only scans need checked OCR before useful text reconstruction.

AI, OCR and document chat

OCR uses local recognition models. Downloadable AI selects summary sentences directly in English, French, Arabic, Chinese and Spanish; it does not translate through English to summarise. Translation supports these languages and German. Check meaning, figures, qualifications and reading order against the source. Local processing does not guarantee an accurate answer.

Chat returns cited source passages. Fast lookup is available without a language-model download; the optional model selects supporting evidence locally. It does not perform reliable arithmetic or provide unrestricted expert advice. An initialised model session can work offline, but fresh sessions can need downloads if browser storage has been cleared or evicted.

Summary and translation results can be read and copied in the workspace or downloaded as formatted PDFs. These reading-copy PDFs preserve their rendered appearance; use the on-screen copy action when you need selectable text.

Signing and archival copies

Visual signatures are page marks. Certificate signing uses an RSA P12/PFX bundle locally, but does not establish issuer trust, check revocation or add a trusted timestamp. Remote signing is not offered. Keep signed originals when making modified copies.

PDF/A conversion uses the selected archival export profile and checks expected output properties. It is not a complete standards validator. Validate the final file with an independent validator such as veraPDF or the receiving archive’s required service, and compare its appearance with the source. Forms, attachments and existing signatures are not retained through the rebuilding conversion.

Device and file limits

Heavy Office and AI engines require substantial downloads and memory. Mobile browsers can suspend background tabs, evict cached resources or run out of memory on large jobs. Keep an original, start with a representative sample and inspect the downloaded result. Read the mobile workflow guide for practical checks.

Found an unexpected result? Contact SandboxPDF with the tool, settings and a non-sensitive description of the problem.