Text Summarizer

When you have a long news story, blog post or report and need the main points quickly, a summary can save time. Paste the text into this tool and it writes a shorter version in its own words, usually a few sentences long, that you can use as a preview or as notes.

The summarisation model runs inside your browser, so confidential documents, draft writing or internal reports are not sent to an online service. Because the model is fairly large, the first run requires a substantial download.

How to use it

1. Paste English text into the input box; a few paragraphs up to a long article works best.
2. On first use, the model downloads, roughly 300 MB, and is cached by the browser for later visits.
3. Choose a summary length if the option is shown, then click summarize.
4. Read the summary alongside the original and check that names, numbers and claims are correct.

How it works

The tool uses DistilBART CNN 6-6 (Xenova/distilbart-cnn-6-6) through Transformers.js. BART is a sequence-to-sequence model from Meta AI: an encoder reads the whole passage, and a decoder writes a new, shorter text one word at a time. DistilBART is a compressed version with six encoder and six decoder layers, and this one was trained on the CNN/Daily Mail dataset of news articles paired with human-written highlights. That is why it tends to produce news-style summaries that reuse key phrases from the source.

Good for

• Previewing a long article before deciding whether to read it in full.
• Creating short study notes from a textbook section or report.
• Drafting an abstract or overview paragraph for your own writing.
• Condensing meeting notes or a long email into the key points.

Limitations

• It works in English only and is tuned for news-style prose; fiction, dialogue, code and bullet lists summarise poorly.
• Very long documents are cut to the model's input limit, so split them into sections and summarise each one.
• Summaries can omit important points or occasionally state something the source does not say, and a compact local model is less capable than large cloud assistants. Always verify before relying on it.

FAQ

More tools