All the essential PDF tools, plus AI that actually understands you.

What each capability actually does

Feature grids are easy to write and hard to trust. This section explains what sits behind each item above: how it works, where it is genuinely good, and where it will disappoint you. If a tool has a weakness, it is described here rather than discovered after you upload a file.

Chat with PDF

The document is extracted to text, chunked, and the relevant chunks are supplied to the model alongside your question, so answers are grounded in the file rather than in the model's general knowledge. It is strongest on questions of the form "where does this document say X", "summarise clause 7 in plain language", or "list every date and obligation mentioned". It is weakest on arithmetic across scattered tables and on scanned pages that have never been through OCR — in that second case, run OCR PDF first and the difference is dramatic. Multilingual questions work because extraction happens before the model sees the text, so a French contract can be interrogated in English.

Smart Translate

Translation is done paragraph by paragraph, preserving block structure, list nesting and table rows rather than flattening the document into one long string. That matters because the usual failure of machine translation on PDFs is not vocabulary — it is losing the shape of the document, so a two-column layout becomes interleaved nonsense. Expect strong results on prose and clause-based legal text, weaker results on dense financial tables with abbreviated headers, and always have a native speaker review anything that will be signed.

Smart Split

Instead of splitting every N pages, the AI reads the document and proposes boundaries where the content changes: a new invoice number, a new chapter heading, a new patient record. You review the suggested split points as thumbnails before anything is written, because an automated boundary you cannot inspect is worse than no automation. For a year of scanned invoices this turns an afternoon into a few minutes; for a continuous narrative document with no structural signals, it has nothing to work with and manual ranges are the better tool.

AI Summarizer and quiz generation

Summaries can be tuned by style — executive brief, detailed outline, plain-language explainer — and by output language. For students, the same pipeline generates practice questions with an answer key, which converts passive reading into active recall. The honest limitation: a summary of a 400-page document is a navigation aid, not a substitute for reading the sections that carry risk. Use it to decide what to read closely, never to decide what you can safely skip in a legal or medical context.

Audio Tutor

Audio Tutor produces a narrated version of the document's substance rather than a robotic read-through of every footnote and page number. It is built for commutes, gym sessions and long drives, and for readers who simply retain more through listening. Files are generated on demand, and shorter documents produce noticeably tighter audio than very long ones.

Low-bandwidth and offline-friendly processing

Every essential tool — merge, split, compress, rotate, crop, watermark, page numbers, image conversion, form filling and signing — executes in your browser using WebAssembly. Nothing is uploaded, which means the tool works on a weak connection, keeps confidential documents on your own device, and returns results as fast as your laptop can compute them. Only the AI tools require a round trip, because inference cannot realistically run on a phone.

The essentials, done properly

• Compression exposes the levers that matter — image downsampling target, re-encoding quality and lossless object cleanup — instead of a single mystery slider.
• Merge preserves bookmarks and page order across files and lets you reorder by thumbnail before export.
• Redaction removes the underlying text objects rather than drawing a black rectangle over them, which is the mistake that has leaked more documents than any other.
• Protect applies AES-256 encryption with separate open and permission passwords.
• Signing supports drawn, typed and uploaded signatures, with fields you place visually on the page.

Who this is built for

Students turning lecture packs into revision material. Small legal teams reviewing contracts without a document-management budget. Accountants splitting a year of statements into something reconcilable. Researchers reading across languages. Freelancers who need one invoice, one signature and one compressed attachment before a deadline, without installing anything. The common thread is people doing serious document work on ordinary hardware and ordinary connections.