Remediate a document

Each file you drop or choose is uploaded as its own job, two at a time. Before anything is sent, this page reads a little of each file in your browser. A PDF that needs a password to open asks for it in the list, and the page checks what you type before it is sent: the service uses the password once to remove the protection and keeps nothing of it, and the remediated file is not password-protected; the report and the job's page say so. A password-protected Word or PowerPoint file (not supported yet) and an old .doc or .ppt are refused in the list with what to do, and are not sent. An encrypted PDF that opens without a password is fine. The service checks each file's own bytes again: anything but a PDF, a Word document (DOCX) or a PowerPoint deck (PPTX) is refused, and so are password-protected Word and PowerPoint files, macro-enabled files and spreadsheets. Size and page limits apply; the report says when a run was page-capped.

It adds structure, alt text, table semantics, reading order, language and metadata, and gives back the same file, plus a report and a review queue for what a person should check. By default nothing visible changes: a PDF is compared with the original pixel by pixel, a Word or PowerPoint file as LibreOffice renders it (Word itself can differ in a few known cases, which the report names). The PDF and Office opt-in modes are off unless you tick them; they can change what a reader sees, and the report names any that changed something, with the pages a PDF mode changed.

Pages and images are sent to AI models through OpenRouter, to zero-data-retention endpoints only. By default a job's files are deleted 30 days after it ends and its content-free records after 90; each job's page shows its dates. Document text is never logged.

A PDF can carry a PDF/UA-1 claim; a DOCX or PPTX never does: its claim reads "remediated against the Microsoft Accessibility Checker rule set, mapped to WCAG 2.1 AA", and only when nothing withholds it. A PDF/UA-1 claim means the machine checks passed (veraPDF's PDF/UA-1 rules, our lint and the pixel comparison); PDF/UA-1 also has checkpoints only a person can confirm, which is what the review queue is for.

A page-capped run never carries a conformance claim. Every claim is backed by the validator runs stored with the job, and the report names anything that withholds it. No conformance claim beyond the report.