AI PDF summarizer that never uploads your file

A 60-page report lands in your inbox an hour before the meeting. A grant call, a contract, a research paper, a policy update – you need the gist now and the details later. The summarizer reads the whole PDF and picks the sentences that cover its main ideas best, then lists them in the order they appear. Each one is tagged with its page number, so you always know where to look for context.

Alongside the summary you get:

• Keywords and key phrases that show up often and across several sections.
• A summary for each section of longer documents, split by the PDF's own headings when it has them, or by groups of pages when it doesn't.
• Quick stats: how many words went in, how many came out, and roughly how much reading time you saved.

You pick the length. Short gives you 3–7 sentences, medium 5–15, long 8–30, and the custom option keeps whatever percentage of sentences you choose. Switch lengths after the first run and the summary updates instantly – no need to process the PDF again.

The tool has two engines and always tells you which one produced the result.

	Extractive summary	Built-in browser AI
How it works	Picks the document's own sentences (TextRank)	Writes a summary in its own words (Gemini Nano in Chrome)
Where it runs	Any modern browser, including phones	Desktop Chrome 138 or later on a capable computer
Languages	Any language	English, Spanish, German, French, Japanese
Accuracy	Quotes word for word, never makes things up	Reads more smoothly, but can get things wrong

The extractive summary builds a graph of sentences: two sentences are linked more strongly the more meaningful words they share. A sentence scores high when many other important sentences point to it – in other words, when it condenses what the document keeps coming back to. Words are reduced to their stems (report, reports and reporting count as one), filler words are ignored, and near-duplicate sentences are dropped so the summary doesn't repeat itself.

The built-in browser AI is Chrome's own on-device language model. If your browser supports it, one click gives you key points or a short TL;DR written in plain language. The first time, Chrome downloads the model (several gigabytes, once); after that it works offline. If your document is in a language the model doesn't accept yet and Chrome's built-in translator is available, the tool translates the important part to English on your device, summarizes it, and translates the result back. The interface flags this, because a double translation can sound a bit stiff. If your browser has no built-in model, you still get the extractive summary.

Splitting text into sentences sounds trivial until a summary starts in the middle of "Fig. 3 shows…" or cuts "Mr. Smith" in half. The summarizer handles the things that trip up simple tools:

• Abbreviations don't end a sentence: e.g., i.e., Mr., Dr., vs., approx., p. 12, No. 5, 3 p.m.
• Decimals, version numbers, URLs and initials (J. Smith) stay intact.
• Bullet points become separate sentences instead of one giant run-on.
• Words hyphenated across a line break are rejoined.
• Running headers, footers and bare page numbers that repeat on every page are left out.

It detects the document's language automatically and works with English and most European languages. It is also tuned for Hungarian, including its abbreviations, dated ordinal numbers and word endings.

A 300-page document won't freeze your browser. The ranking runs in a background thread (a Web Worker) and works section by section, so it never has to compare every sentence with every other one. A progress bar shows where text extraction is, and you can cancel at any time. If you only need part of the document, enter the pages, for example 1-20, 35.

A scanned PDF contains pictures of pages, not text. The tool tells you how many pages are image-only and lets you turn on OCR (with English and Hungarian language data). Recognition also runs on your device and takes a few seconds per page. In mixed documents only the image-only pages are recognized; the rest is read directly.

The summary is created entirely in your browser. The PDF is never uploaded to a server, there's no sign-up, and you don't need an account with any AI provider. That matters for confidential material – internal reports, proposals, medical or legal paperwork. The only things your browser downloads are the app code and, if you ask for OCR, the language data. Chrome's built-in AI model is downloaded and stored by Chrome itself.

Tips for a better summary:

• Use the page numbers. If a sentence matters, jump to that page and read around it.
• For long documents, read the section summaries. They often tell you more than the overall summary, because every section gets a voice.
• Too general? Switch to long, or narrow the page range to one chapter.
• Before any decision involving a contract, a deadline or an amount, read the original. The extract quotes the text exactly, but the selection is automatic and can miss a detail.