Toolvore

AI Summarizer

Condense long text into a summary or key bullet points.

This tool sends the text you enter to an AI provider to generate a response. Don't paste secrets or personal data. No files are uploaded.

Long text goes in and a short version comes out — one sentence, a short paragraph, or three or five key bullet points, chosen from a dropdown before you run it.

How to use it

  1. 1Paste the full text into the box; there is no file upload, so articles and documents have to be copied in as text.
  2. 2Pick the summary length: One sentence, A short paragraph, 3 key bullet points or 5 key bullet points.
  3. 3Press Summarize and copy the result when it stops streaming.

Example

Input
A 1,200-word news article, with '3 key bullet points' selected
Output
Three bullets covering the article's main claims, drawn only from the text you supplied.

The model is instructed not to add anything that is not in the source, so a gap in the article stays a gap in the summary.

What happens to your data

Condensing happens at the model, never in the page, which means the entire document crosses the network — there is no local step that picks out sentences and sends a sample. Our /api/ai route counts the characters first and rejects anything over 40,000 with 'Input is too long', so an oversized paste reaches our server but is turned away before it reaches Anthropic; anything under the limit is relayed straight through. That makes this the wrong place for board papers, medical letters or unreleased material, and it is worth reading the summary against the source before quoting it, since compression is exactly where a qualifier or a caveat gets dropped.

Last updated August 2026

Somebody has sent you forty pages and asked what you think by Friday. Or a thread ran to sixty messages while you were away, or a supplier's terms arrived as a wall of clauses. The task is not really reading — it is working out what is in there and whether it deserves your attention.

Two different jobs get called summarising, and it pays to know which one you have. The first is compression: keep the shape of the argument, lose the length, so a reader of the short version knows roughly what the long one said. The second is answering a question you already hold — does this contract auto-renew, did anyone approve the budget. Summarising is poor at the second, because it keeps what the source made prominent rather than what matters to you. When you know what you are after, searching the text for the term beats condensing it.

The input is text you have pasted, which decides more than it sounds like it does. Anything whose meaning lived in its layout — a table, a slide deck, a spreadsheet — arrives flattened, and a table read in reading order becomes a run of numbers with the headings adrift near the top.

The mistake worth naming is treating the short version as a substitute where the details are the point. On a contract, a diagnosis or anything you will be held to, a summary tells you where to read closely. It does not save you from reading.

How it works

Toolvore wraps a single line around what you paste — Summarize the following text as three key bullet points, or whichever length you picked — and posts the pair to this site's own /api/ai route, which adds a fixed instruction of its own (capture the essential points faithfully and neutrally, add nothing that is not in the source) and streams the reply back a few words at a time. That is why text appears before the answer is finished, and why the Copy button turns up only once it stops. Two ceilings are worth knowing: the reply is capped at 2,000 tokens, and one that reaches the cap ends with a bracketed note rather than trailing off; and requests are limited to six a minute from one address. The real weakness is that nothing compares the result against the source — an omission is silent, and a section that got no bullet looks exactly like a section that had nothing in it.

Common use cases

  • Deciding whether a long report needs reading in full before a meeting
  • Reducing a sixty-message thread to a few points for a handover
  • Triaging research papers by the claims they actually make
  • Condensing meeting notes for the people who were not there
  • Drafting an abstract for a document you have written
  • Finding the common themes in a batch of customer reviews

Frequently asked questions

How long should a summary be?

The length is set by what the reader will do with it, not by the size of the source. Journal abstracts run to roughly 150 to 250 words whatever the paper's length; an executive summary is conventionally a page. As a ratio, once you drop under about a tenth of the original you stop describing the argument and start naming the topic. Watch what each length can carry, too: a paragraph is the only shape with room for a because, so where a source's value is its reasoning, a single sentence throws exactly that away.

What is the difference between an extractive and an abstractive summary?

An extractive summary lifts whole sentences out of the source and sets them side by side. Every word is verbatim, which makes it easy to check, and it reads bumpily wherever the chosen sentences did not originally follow one another. An abstractive summary writes new sentences describing what the source said, which reads far better and is what happens here — the text goes to a language model that produces prose, not through a scoring pass that picks lines. The trade is verifiability: no sentence can be found in the original by searching for it.

Are AI summaries accurate?

Reasonably, and they go wrong in a particular direction. Outright invention is the rarer failure, especially under an instruction to add nothing absent from the source. The common one is omission and flattening: a section that got no bullet, a hedged finding restated as a firm one, a minority objection given the same weight as the consensus. None of that announces itself, and a summary that dropped the paragraph you most needed looks exactly like one that did not. Run your eye down the source's headings and ask which of them left no trace.

Is it private to summarise a document here?

Your text leaves your machine, so this is the wrong place for anything confidential. It is posted to this site's /api/ai route, which checks the request against a per-address rate limit, refuses anything longer than 40,000 characters before going further, and relays the rest to Anthropic's API using a key held on the server — that key and the fixed instruction never reach your browser. Pressing Stop cancels the stream you are receiving, not the copy already sent. Board papers, patient letters and anything under an NDA belong somewhere that is not a web form.

How do I summarise a PDF, a web page or a video?

Get the words out first, because the box takes typed text and nothing else. Most PDF viewers let you select and copy the text, and a PDF-to-text tool helps where the columns fight you; a scanned page has no text to select at all and needs OCR before anything can read it. Web pages copy far more cleanly from a browser's reader view, which strips the navigation and the cookie notice you would otherwise be summarising too. Video and audio need a transcript, and platform captions work once the timestamps come out.

What do I do with a document too long to paste in one go?

Split it and summarise in two passes. Cut at the document's own seams — chapters, sections, dated entries — and never mid-argument, since a chunk beginning halfway through a point will summarise the second half as though it were the whole thing. Summarise each part, paste those summaries together, then summarise the result. Be clear about the cost: the second pass sees only the first pass's output, so anything dropped early is gone for good, and a detail that mattered in one section gets levelled out against the rest of the document.

Should a summary be bullet points or a paragraph?

Bullets suit sources that were already lists — minutes, release notes, a set of findings, anything whose items do not depend on one another. They are the wrong shape for an argument, because a bulleted list has no way of saying therefore or unless, and a reader can mentally reorder the points without noticing the reasoning has broken. A paragraph keeps those joins intact. Three bullets or five is a decision about the reader: three is what somebody scans inside a message, five is what survives being pasted into a working document.

Can I tell it which parts to focus on?

Not through the interface — the only control is the length dropdown, and the instruction sent alongside your text is fixed. Typing an aside into the box, along the lines of ignore the first section, is unlikely to work either, and that is deliberate. The instruction held on the server tells the model that everything you supply is material to be processed rather than directions to be followed, including anything phrased as a command to an AI. Pasted text usually came from somewhere else, and a document that could redirect the tool by containing the right sentence would be a real problem.