What PDF Guard does locally
PDF Guard does not transmit the working document to a hosted language model. The browser reads PDF objects, renders pages when needed and runs an OCR recognition model on the device. After text is available, local rules identify likely abbreviations, labelled values and checklist matches. This focused approach makes the data path understandable and keeps sensitive document material out of a third-party prompt.
A local-first data path
The path is deliberately short: your PDF goes from your file picker into browser memory, the browser performs the analysis, and the result appears in the same tab. Static application files such as JavaScript, the PDF reader and the English OCR model are downloaded from the website, but your document does not travel back with those requests. There is no document account, server-side analysis job, upload bucket or permanent report store in the current architecture.
Useful work without a general-purpose cloud model
A focused local checker can handle common review preparation well. It can detect whether a PDF contains searchable text, OCR clear scans, compare text with a list of expected items, locate dates and reference labels, and explain a maintained set of common abbreviations. Unknown acronyms are flagged for contextual confirmation rather than invented. The result is intended to help a person navigate a document faster, not to replace professional judgement or provide an authoritative interpretation.
Understand the boundary
Privacy depends on implementation, not a slogan. PDF Guard does not send document bytes, OCR text, checklist text or report values to analytics. Anonymous page and feature events may be counted, but they contain only safe categories such as page type and file-size range. If a future feature sends text to an external AI service, it must be clearly labelled before use and documented here. The current local checker does not include that type of external model call.
When local processing is the better fit
Local analysis is particularly useful for draft procedures, administrative forms, personal records and early quality checks where uploading the original is undesirable. It also works without creating a user profile. Device performance matters: long image-heavy PDFs can take time, and mobile browsers have tighter memory limits. For very large files, split the document into smaller sections and review each local report against the original.