Summarizer
Offline small-model summary.
Loading tool…
How automatic summarisation works
There are two families of summarisation. Extractive summarisation scores the sentences already in a document and returns the highest-ranking ones — classic approaches use word frequency, position, and graph algorithms such as TextRank, which treats sentences as nodes and their similarity as edges. Abstractive summarisation generates new sentences that paraphrase the source, which reads better and requires a large language model to do it.
Extractive methods have a decisive practical advantage: they never invent anything. Every sentence in the output appeared verbatim in the input, so there is no risk of a fabricated claim. The trade-off is that the result can read as a slightly disjointed set of highlights rather than flowing prose. For skimming a long document to decide whether it deserves a full read, that is usually the right trade.
Why we built this tool
People summarise to triage: a long report before a meeting, an article to see whether it answers a question, meeting notes into action points, or research material into something reviewable.
The documents in question are usually confidential — internal reports, contracts, research under embargo, client correspondence. Pasting them into an online summariser sends the entire text to a third party, and with hosted AI services it may be retained or used for training. Ours runs the analysis in your browser with JavaScript, so the document never leaves your device. That constrains us to extractive methods rather than a hosted model, and we consider the privacy worth the trade.
Tips and common mistakes
- Read the summary against the source before relying on it; extraction can drop the one qualifying sentence that mattered.
- Well-structured source text with clear topic sentences summarises far better than rambling prose.
- Aim for 10 to 20 per cent of the original length as a starting point.
- Use the output as a skim aid, not as a replacement for reading anything you will be held to.
Open Summarizer with a link
Summarizer can be handed its input by the link that opens it, so the page arrives with the work already done. This is meant for assistants and scripts: if a chatbot, an editor extension, or a command-line agent already holds your content, it can build a link instead of asking you to copy and paste into a box.
The payload rides in the fragment — the part of a URL after the #. Browsers never send the fragment to a server, so anything handed over this way stays on your device exactly as a dropped file would. A link that uses the query string instead is rewritten into the fragment by the page before any analytics or ad script can read the address.
Prefix any field with gz — #gztext= — to pass it as gzip-compressed base64url, which fits roughly five to ten times more into the same link. Links stay reliable up to about 8,000 characters; past that, chat apps and address bars start truncating them, and the page says so rather than loading half a document.
One exception to the privacy rule: ?url= tells the page to download the input from an address. That request goes from your browser straight to that server — not through ours — but it does leave your device, and it only works if the server allows cross-origin requests.
#text=— the text to work on
Example: https://mygadgets.ink/tools/summarizer/#text=...
Quick start: using Summarizer
- Paste the text you want condensed into the input above.
- Choose how long the summary should be, as a length or a sentence count.
- Run the summarisation and read the extracted sentences.
- Check them against the source for anything important that was dropped.
- Copy the summary, treating it as a starting point rather than a finished piece.
Measure the reduction with Word Counter, and pull source text out of documents using Extract Text from PDF.