Summarize a PDF
Key sentences from every section, an optional written summary, and a slide outline.
By opening this website or using its tools, you agree to our Terms and Privacy Policy.
or drag it here
PDF — nothing is uploaded
Your file never leaves your browser during processing
How it works
To summarize a PDF, drop it onto this page — the text is extracted in your browser, the most representative sentences of each section are picked out with no model at all, and where your browser provides an on-device model it can also write a summary and a slide outline. The file is never uploaded.
The document is read in your browser with pdf.js, split into sections on page boundaries, and ranked. The key sentences you get first are extractive: each sentence is scored on the words it shares with the rest of its section, with a bonus for appearing early, and the best few are shown in reading order with their page numbers. Every sentence there is a sentence the PDF actually contains — nothing is paraphrased, and no model is involved, so it works the same in every browser including on a phone.
The written summary above that is optional and runs on your browser’s own built-in summariser, which Chrome and Edge ship on desktop. The interesting constraint is not the model’s quality but its context window: it holds a few thousand tokens, and a twenty-page report does not. So the document is summarised section by section and those summaries are then summarised together — map, then reduce. When that second pass happens the tool says so, because a summary of summaries compresses the middle of a long document twice and the section summaries underneath are closer to what the document actually says.
The slide outline goes through a different API for a real reason. Summarisation returns prose; an outline is structured output, so it is generated through the Prompt API with a schema constraining the reply to titled slides with bullets, and the reply is parsed and checked rather than trusted — a model that misses the shape is reported, not rendered as an empty deck. The outline is built from the condensed text, never from the whole document.
Scanned pages have no text layer, so nothing from them reaches any of this. Those pages are named individually and can be read with OCR on a second pass, which is opt-in because it downloads a recognition engine and is far slower. Nothing here calls a server: the summariser and the outline model are downloaded and managed by your browser, and your document stays in the tab.
FAQ
How do I summarize a PDF for free?
Drop it onto this page. The key sentences of each section are extracted in your browser with no model and no account, and if your browser provides an on-device model you can also generate a written summary and a slide outline. Nothing is uploaded and there is no sign-up.
Does this work without AI?
Yes, and that is the default. The key-sentence pass is a term-frequency ranking that runs in every browser, quoting real sentences from the document. The generated summary is an extra layer on top where a browser provides a model.
Why is my summary split into sections?
Because the browser’s built-in model has a small context window — a few thousand tokens — and a long PDF does not fit. Sections are summarised separately and then combined, and the tool tells you when that second pass ran.
How long a PDF can it handle?
There is no page limit imposed here; the practical ceiling is your tab’s memory for the extraction, and time for the summary, since every section is a separate model call. A very long document takes a while and the section progress is shown as it goes.
My PDF is a scan and nothing came out. Why?
A scan is a picture of a page with no text layer behind it. Those pages are named in the result, and can be read with OCR in a second pass — that is offered as a choice because it downloads a recognition engine and is much slower than reading a normal PDF.
Is my PDF uploaded or used for training?
Neither. Extraction, ranking, summarisation, and the outline all happen in your browser, on a model your browser downloads and manages. The network counter under the tool stays at zero bytes uploaded while your document is processed.