AI PDF translator that keeps formatting and masks sensitive data locally

Upload a text-based PDF, choose the source and target languages, then review the sensitive values found in the extracted text. After you confirm the review list, the browser replaces those values with stable placeholders and sends the masked paragraphs for translation.

When the translation returns, placeholder integrity is checked, real values are restored locally, the source text is permanently removed from the PDF, and translated text is inserted at the original positions.

Page size, images, figures and the surrounding page structure stay in place. Because translated text can be longer or shorter, the result closely preserves the layout rather than promising a pixel-identical copy; long passages may use a smaller font.

Scanned PDFs, encrypted files, rotated text and vertical writing are not supported. If text cannot be selected in the original PDF, this workflow cannot translate it because OCR is outside the supported scope.

The PDF file itself is parsed and rebuilt locally. Confirmed sensitive entities become placeholders such as ⟦NAME_1⟧ and ⟦BANK_1⟧ before translation. BYOK sends masked text and the API key held in the current tab directly to your chosen OpenAI-compatible endpoint; managed mode relays masked text through Veil Translate to a fixed Cloudflare Workers AI model and never receives a user key.

This is not encrypted translation: the de-identified body text remains readable to the provider handling the request. Review the detected values and add any missed sensitive information before starting.

It is suited to selectable-text reports, contracts, financial documents and internal materials that need a first-pass translation with reduced exposure of confirmed sensitive values. Always review high-stakes legal or financial translations before relying on them.