Paper Reader

Ask about the sentence you cannot follow.

Select a passage. Ask about it in the margin. The question stays pinned to those exact words while you read on.

Open a paper now

No account, nothing to install. The demo gives you 5 questions on a real paper, and saves nothing.

An illustration, drawn here in HTML and CSS: it is not a screenshot. Everything inside it is real, though. The answer is the product's own, asked of this passage at its default settings and pasted back word for word — including the last clause, where it declines to state a result the passage does not give. It cost about a tenth of a rupee, which the reader is shown beside it. The paper is real too: Michelle E. Stepan, Taylor M. Dehnke, Kimberly M. Fenn. PLOS ONE 12(9): e0182907 (2017). doi:10.1371/journal.pone.0182907. Reproduced under the Creative Commons Attribution 4.0 International licence (CC BY 4.0), creativecommons.org/licenses/by/4.0/.

The margin

The question goes beside the words it is about.

You hit a clause that assumes a background nobody ever taught you. Select it. A thread opens next to the passage — you ask there, and the answer arrives there, rather than in a separate chat window you keep translating back to the page.

Above the thread you say how much it should read (just your selection, or the whole paper), how much of the field you already know, and whether you want the answer brief or detailed.

Figures, tables, equations

A formula is not text. So you click it instead of selecting it.

A displayed equation does not come out of a PDF as an equation. It comes out as rubble: the limits of a sum landing beside it instead of above and below it, an integrand collapsing into floor brackets. Paste that into a chat window and you get an answer about gibberish.

So Paper Reader finds the figures, tables and equations while it reads the document and marks each with corner brackets. Click one and the question carries the region of the page itself — the formula as printed, the axes and the caption, the whole table with its header row.

An illustration. The lower half is a real extraction, copied out of a real PDF; the upper half is that same equation drawn here as the page sets it, with the integrand left off. Neither half is a screenshot of the application.

It stays put

Come back on Friday. The thread is still on the sentence.

Threads are anchored to the passage, not to a session. Scroll away, close the tab, sign in from another machine: the question is still hanging off the same words with everything you asked underneath it. Ask a follow-up without setting the scene again — the context is what the thread is attached to.

An illustration of one thread across two sittings; not a screenshot. The sentence is again from Stepan, Dehnke & Fenn, PLOS ONE 12(9): e0182907 (2017), doi:10.1371/journal.pone.0182907, reproduced under CC BY 4.0, creativecommons.org/licenses/by/4.0/.

How it works

Three steps, and no setting-up.

  1. Open a paper

    Upload one, paste a link, or start from the open-access papers offered while your library is empty.

  2. Select what you cannot follow

    A sentence, or a figure, a table or an equation by its corner brackets.

  3. Ask, and read on

    The answer lands in the margin beside it, and stays there.

PDFHTMLMarkdown — the three things it reads.

What it costs

Three ways in. One of them never asks you for anything.

The full pricing, and the honest trade-off of using your own key.

Open a paper and select the sentence.

Open a paper now

5 questions, no account, nothing saved.