A text summarizer that runs on your device
This text summarizer turns a long piece of writing into something you can read in seconds: a few bullet points, a short paragraph, a one-line TL;DR or a headline. Paste an essay, a report, a policy, a chapter of notes or a long message — or open a .txt, Markdown or Word (.docx) file — choose the format and the length, and press Summarize.
Where it can, the summary is written by an AI model that runs inside your browser, so the text you paste is not uploaded. Recent desktop Chrome has a summarizer built in; on other browsers the page can download an open-source model (IBM’s Granite 4.0, Apache-2.0 licence) once, with your permission, and run it on your own processor or graphics chip. A third, optional engine picks the most important sentences of your text on our server instead — it never rewrites anything.
How to summarize text
- Paste your text into the box, or press Open .txt, .md or .docx (you can also drop a file onto the page). The word count updates as you type.
- Choose a Format — Key points, Paragraph, TL;DR or Headline — and a Length. The hint under the length says exactly what you will get, for example “5 bullet points”.
- Check the AI engine line: “Automatic” uses your browser’s built-in AI if it has one, otherwise the open model on this device. Change lets you pick.
- Press Summarize. If a model has to be downloaded first, the page asks you and shows its size; the summary then appears word by word. Stop ends it at once and keeps what was written so far.
- Copy the summary, or download it as plain text or Markdown. The panel underneath shows words before and after and which engine wrote it.
Text Summarizer features
Three engines, one honest label
Browser AI, an open model on your device, or server-side key-sentence picking — the badge always names the engine that produced the result.
Four formats, three lengths
Key points (3, 5 or 7 bullets), a paragraph (about 3, 5 or 8 sentences), a TL;DR of one to three sentences, or a single headline.
Long texts in parts
Text that does not fit the model in one go is split at paragraph and sentence boundaries, summarised part by part, then combined into one summary.
Asks before any download
The open model is about 340–550 MB depending on your device and is downloaded only after you agree; your browser then keeps it, so later summaries start quickly.
Reads your files locally
Text, Markdown, HTML and Word .docx files are read in the browser, never uploaded.
Which engine should you use?
| Engine | Where it runs | Download | Good to know |
|---|---|---|---|
| Browser’s built-in AI | Inside Chrome (Gemini Nano) | Managed by Chrome, once, shared by all sites | Desktop Chrome 138 or later; needs about 22 GB of free disk and either a graphics chip with more than 4 GB of memory or 16 GB of RAM. Summarizes English, Spanish, Japanese, German and French. |
| Open model on this device | Your graphics chip (WebGPU) or processor (WebAssembly) | 338 MB (WebGPU) or 553 MB, asks first; Granite 4.0 1B is 1.2 GB | Works in most modern browsers. Processor mode is noticeably slower; a phone will struggle with long texts. |
| Key sentences (server) | This site’s server with TypeSafe’s Jev model | None | Sends your text’s sentences when you press the button. Picks the most important sentences; it does not rewrite or shorten them. |
What an AI summary can and cannot do
An abstractive summary — one written in new words — is quick to read, but a small model can still get things wrong: it may drop a qualifier (“about”, “in the first year”), merge two numbers, or state something more confidently than the original. Models small enough for a laptop are private, but less capable than large cloud models. Treat the summary as a reading aid and check any figure, name or date against your text before you quote it.
The key-sentence engine works differently. It scores every sentence for importance and keeps the best ones in their original order, word for word, so it cannot invent anything — but the result can read a little choppy, and a pronoun such as “it” may point to a sentence that was left out. For PDFs, the PDF Summarizer uses the same approach with page numbers.
Tips for a better summary
- Remove material you do not want summarised, such as reference lists or repeated disclaimers — small models do best with clean input.
- Use Key points for reports and notes, Paragraph when you need something to paste into an email, and Headline to title a document or a post.
- For web pages, use the Article Summarizer, which fetches the page and strips menus and ads first. Meeting transcripts have their own Meeting Summarizer that also lists action items.
- For detailed counts, use the Word Counter; to check how easy your text is to read, the Readability Checker.
Text Summarizer FAQ
Is this text summarizer free?
Yes. There is no sign-up and no word limit set by us. The on-device engines run on your own hardware, so the only limit is what your device can process in reasonable time.
Is my text uploaded?
Not with the browser’s built-in AI or the open on-device model — the text stays in your browser and only the model files are downloaded. If you choose “Key sentences (server)”, the sentences are sent when you press the button, scored and not stored.
Why does it want to download a model?
Your browser has no built-in summarizer, so the page can run IBM’s open Granite 4.0 model locally instead. It asks first and shows the size (338–553 MB for the default model). Your browser caches it, so it is downloaded only once.
How accurate are AI summaries?
Usually good at capturing the main points of clear prose, but small models can misstate numbers or miss nuance. Always check important details against the original. The key-sentence engine only quotes your own sentences, so it cannot introduce errors — but it cannot rephrase either.
Can it summarize very long text?
Yes. Text that does not fit the model in one go is summarised in parts and the part summaries are combined. That takes longer, especially on the processor, and very long documents lose some detail.
Which languages work?
Chrome’s built-in summarizer handles English, Spanish, Japanese, German and French. Granite 4.0 is trained mainly on English plus German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch and Chinese; other languages may give weak results.