Paste a long piece and get the sentences that carry most of it, with a slider for how many you want. This is extractive summarising — it picks sentences out of your text rather than writing new ones — so every word in the summary is a word you wrote. Sentences score on how often their significant words appear across the whole document, divided by length so that long sentences do not win simply by containing more words. They come back in the order they appeared, because a summary sorted by relevance reads as nonsense: a conclusion before its premise is incoherent however relevant both sentences are. No model, no network, nothing uploaded, and it works offline.
| This tool | An AI summariser | |
|---|---|---|
| Writes new sentences | No | Yes |
| Can invent facts | No | Sometimes |
| Needs a network | No | Usually |
| Your text leaves the device | No | Usually |
| Condenses two ideas into one | No | Yes |
No. It scores sentences by word frequency and picks the highest. There is no model, no API call and no network use at all, which is why it is instant, works offline, and why your text never leaves the page.
No. Every sentence in the summary is taken verbatim from your text. That is a genuine limitation — it cannot condense two sentences into one — but it also means it cannot invent anything that was not there, which generative summarisers sometimes do.
Because sequence carries meaning. Returning them ranked by score produces something that reads as gibberish even when each sentence is individually apt, since arguments depend on order. Original order keeps the summary readable.
Three to five works for most articles. Fewer risks losing the argument; more stops being a summary. Move the slider and read — the right number is obvious once you see it.
Frequency scoring has no understanding of meaning. A sentence packed with the document's most common significant words scores highly even if it is an aside. It works well on focused writing and less well on text that wanders between topics.
A run ending in a full stop, question mark or exclamation mark. Common abbreviations like Dr. and e.g. are handled so they do not split a sentence in half.
This is extractive summarising: it selects sentences from your text rather than writing new ones. Each word that is not a stop word contributes its frequency across the whole document, a sentence scores as the average of its words' frequencies, and the highest-scoring sentences are kept. Dividing by sentence length rather than summing is the detail that makes it work — an unnormalised total simply picks the longest sentences, since a longer sentence contains more words and therefore more frequency to add up. The other decision that matters is returning the selected sentences in their original document order rather than by score. A ranked summary reads as nonsense even when every individual sentence is well chosen, because an argument depends on sequence: a conclusion arriving before its premise is incoherent regardless of how relevant both are. Sentence splitting guards against abbreviations ending in a full stop, since cutting 'Dr. Smith' in half would produce a fragment that scores oddly and reads worse. Nothing is generated and nothing is transmitted — there is no model and no network call, which is why it works offline and instantly, and also why it cannot paraphrase.