Facts drawn from the ISO specifications that define the format, followed by the workflow patterns we observe building browser-based tools. Everything on this page is either sourced to a standard or labelled as an observation.
Each of these can be verified directly in the published standards.
PDF was published by Adobe starting in 1993, and became an open ISO standard in 2008 as ISO 32000-1, corresponding to PDF 1.7. The specification defines the object model, page tree, content streams, filters, fonts, and security handlers.
PDF 2.0 removes a number of deprecated features, tightens the specification, and adds capabilities including improved support for tagged content and accessibility. Files in the wild are still overwhelmingly PDF 1.x.
Compressed object streams and cross-reference streams are what make lossless optimisation possible: they let a producer store the document's structure in compressed form rather than as plain-text tables. Rewriting a file with object streams is the main lossless size reduction available.
PDF's standard lossless filter is Flate, which is the same Deflate algorithm used in ZIP archives. Content streams and embedded images may use it. Recompressing streams with Flate changes nothing about how the document renders.
Standard PDF encryption is defined by security handler revision. Revisions 2 and 3 use RC4 with 40-bit or 128-bit keys and should be considered broken. Revision 6, using AES-256, is the current strong option. A file protected with revision 6 opens in any conforming reader.
PDF/A is a constrained subset of PDF intended for archival storage. It requires all fonts to be embedded and prohibits features that may not render in the future, such as encryption, JavaScript, and external content references. Conformance is verified by a validator, not by a re-save.
PDF/UA defines requirements for universally accessible PDF, including tagged content, a logical reading order, and alternative text for images. It is a separate conformance regime from PDF/A, and a document can be one, both, or neither.
PDF/X is a print-oriented subset designed to make files reliably reproducible in commercial printing. It is unrelated to the PdfPix brand name despite the similar spelling, and it is a common source of confusion in search results.
The format is specified openly and implemented openly. Mozilla's PDF.js renders pages in the browser, pdf-lib reads and writes documents in JavaScript, and Poppler and MuPDF are mature native implementations. This is why a browser can manipulate a PDF locally without a server.
Qualitative assessments rather than forecasts. Each is labelled with how settled we consider it.
WebAssembly lets compiled C, C++, and Rust code run in a browser at near-native speed, and WebGPU exposes the GPU to web pages. The practical effect for document work is that parsing, rewriting, and rasterising a PDF no longer requires a server. This is the single biggest change in the last few years, and it is what makes local-first document tools viable rather than aspirational.
Status: Established
Users increasingly distinguish between tools that promise to delete an upload and tools that never receive one. The second is an architectural guarantee and the first is a policy, and the difference is verifiable in a browser's network panel. We expect this distinction to become a standard part of how document tools are evaluated.
Status: Established
A model has to read text to summarise or translate it, so any AI feature either runs on the device or transmits the text. Both are legitimate designs, but they are different products. The emerging norm is that a tool should state plainly which one it uses, and users should expect that answer to be testable.
Status: Emerging
The European Accessibility Act applies to a wide range of digital products and services, and accessibility litigation continues in several jurisdictions. Tagged PDF with a correct reading order and alternative text is moving from good practice to obligation for many organisations, and it is a genuinely hard problem to retrofit.
Status: Emerging
Merging, splitting, rotating, cropping, and stamping are solved problems with mature open-source libraries behind them. As these operations become free and local everywhere, the differentiation moves to the harder cases: high-fidelity conversion, genuine redaction, accessibility remediation, and archival conformance.
Status: Established
These are patterns we have seen repeatedly. They are qualitative, and they are offered as things to check rather than as measurements.
Because structural PDF work is a solved problem with mature open libraries behind it, it belongs on the client. That is the constraint the tools are built against: merge, split, organise, rotate, crop, watermark, paginate, sign, and the image and text conversions run in your browser and never transmit a document. Where a capability genuinely needs a model — summarisation, translation, and the text-cleanup step in OCR — the feature says so rather than inheriting a blanket privacy claim.