PDF, DOCX, EPUB, ODT, TXT, MD, HTML, or Image
or use the buttons above
Please don't submit sensitive personal data.
Please don't paste sensitive personal data.
Space Play / Pause
← → Step backward / forward
↑ ↓ Adjust word stream speed
Hold Shift Slow the word stream while held, back to your speed on release
Ctrl / ⌘ F Search this document
Read Aloud narrates the text while keeping the word stream synchronized. Voice speed and visual word-stream speed are controlled separately.
Section estimates the time remaining in the current section. Full-Text estimates the time remaining in the entire loaded file.
Pacing varies how long each word stays on screen according to how common that word is. Ordinary words like the and was pass quickly; unfamiliar and technical words are held longer, so a term you are meeting for the first time gets the extra moment it needs instead of going by at the same rate as everything else.
There are three settings. Even gives every word the same length of time. Subtle, the default, applies about half the variation. Natural applies it in full, where the longest-held words stay on screen roughly three times as long as the shortest.
Whichever you pick, the words-per-minute you set does not change. The time is redistributed between words rather than added to them, so a document takes the same length of time to read at every setting — only the rhythm differs.
Pacing takes account of the document itself, not only general English. A term that runs throughout a paper counts as part of that paper's vocabulary, so words particular to your field pass at full speed rather than being held as though they were unfamiliar.
The brief pauses at sentence and clause endings are separate from this and stay active at every setting, including Even. Pacing applies to silent reading only; Read Aloud follows the voice's own timing.
The Search tab finds any text in the open document. Results are grouped by section, and clicking a result jumps the word stream to that point without losing your place. Search matches partial words, so "ion" also finds "solution". Press Ctrl / ⌘ F from anywhere to jump straight to it.
Use Contents and Text to move around the document. Clicking text or a contents item updates the word stream position.
Select any passage in the Text tab and a small toolbar appears. Save to Notes keeps it as a highlight, tagged with the section it came from and the point in the document where it sits, so it appears in Notes and in the Study panel's saved items and stays findable later.
The other button on that toolbar, Split here, starts a new paragraph immediately before your selection. Use it where extraction has run two things together — a heading absorbed into the paragraph beneath it, or two sections joined across a page break. Splitting affects the reading stream, so the pieces can then be moved or removed independently in Arrange.
Make heading, on the same toolbar, turns the selected words into a heading. It appears in Contents at that point in the document and starts a new section, so Study treats what follows as its own section too.
Use it where a document's own structure did not come through: a paper with no table of contents and no detected headings, or one where only some sections were found. Selecting part of a paragraph splits it, so the heading is the words you chose rather than the whole paragraph.
Selecting inside any heading changes the button to Remove heading. It leaves the Contents and the text reads on as body. This works on headings Gwion detected as well as ones you made — useful where a displayed equation, a caption or a stray line has been taken for a heading.
Arrange, at the top of the Text tab, turns the text into blocks you can reorder. Hovering a paragraph shows arrows to move it up or down and a cross to take it out of the reading stream. Removed paragraphs are not deleted — they stay in place marked removed, greyed out, with a Restore button, and they are skipped by narration and by the word stream.
This is for fixing what extraction got wrong on a particular document: a running header spliced into a sentence, a caption landing mid-paragraph, a footnote interrupting the prose. Undo steps back one edit at a time and Revert restores the original extracted order.
Your edits are saved per document and are anchored to the text itself rather than to positions, so they survive re-extraction — including a Rebuild — and are still there when you reopen the document.
Study mode is a three-part workspace. The left panel navigates and tracks progress, the center panel presents the document as study material, and the right panel holds your tools.
Left navigates by Outline, Concepts, Key Points, or Saved items. Center shows the document as Key Points, a Simplified rewrite, or a Concept Map. Right holds the Study Coach, Quick Tools, and tabs for Notes, Practice, and Figures.
The center panel's select bar — Explain, Simplify, Quiz This, Flashcard, Save Highlight — stays disabled until you select text, then acts on that selection.
The right panel's Quick Tools — Explain, Simplify, Memory Anchor — use your selection if you have one, and otherwise fall back to the current section. Quiz Section always uses the current section.
The Practice tab generates a quiz or flashcards for what you're studying, drawing on the section's figures as well as its text, and Figures shows figures related to your position. Add to Review Sheet saves the current material to revisit later in Notes.
Quizzes and flashcards are for self-checking: reveal each answer or flip each card and judge yourself. Gwion does not grade or score your answers — the only scoring anywhere is what you report to the coach yourself.
After studying a section, log the session. The coach asks four questions about what you did — not about the material, since Quiz and Flashcards already cover that — and returns two things. Understanding is how that session went. Recall is how long it is likely to last, and it fades on its own from there until you log another session, so a section left alone will eventually be marked as due.
How you studied matters as much as how it felt. Recalling something without looking builds more durable memory than rereading it does, even though rereading feels more fluent at the time. So the same answers after a reading pass produce a shorter interval than after a quiz. Denser sections — more equations, more figures — also fade faster, so they come up for review sooner.
When other sections in the document have come up for review, the coach shows the first of them under Due, with a count if there are more. That is review order, not document order — a section due for review may be behind you as easily as ahead. Clicking it takes you there.
How much to trust it. The specific numbers are a rough guide, not a measurement — nothing here is calibrated against your results or anyone else's. Read the output as a comparison, not a reading: “this section is shakier than that one” is meaningful, “holds about a week” is a rough shape rather than a deadline. That is why Gwion says solid, shaky, and fading rather than a percentage, and gives horizons in weeks and months rather than exact days.
The coach is a prompt to review, not a verdict on what you know. If it says a section is solid and you are not confident, trust yourself over it.
Some PDFs carry a text layer that is wrong rather than missing — a scan whose characters were guessed at, or a paper whose equations arrive as scrambled symbols. When Gwion can tell it is looking at a document like that, it offers to re-read the pages in the Equations panel.
You pick the pages, at the equations. Every equation in the Equations panel carries a Re-read page button. Pressing it selects that page rather than starting anything, so you can look through the crops, mark the ones whose text does not match the picture, and see the total before you commit. A bar at the foot of the panel shows how many pages you have chosen and roughly how long they will take; nothing is read until you press it.
One section at a time. Once you select a page, the buttons on pages outside that section dim — a page that reads badly usually means its neighbours do too. Clear the selection to choose somewhere else. This works on a document of any length.
Whole documents. When a document has almost no readable text at all — a scan with no usable text layer — every page needs re-reading, and Gwion offers that up front. It is part of the paid plan; a free allowance covers a few pages, enough to rescue the equations in an article. Single pages with equations on them stay open to everyone. The whole-document offer is limited to documents of about forty pages.
What it does. Each page you picked is rendered as an image and read back as text, and its equations come back as real mathematics rather than loose symbols. That page's text then replaces what was there before, so the reading stream, the Text tab and narration all pick up the corrected version. Pages you did not pick are untouched.
What it costs, and how to undo it. Pages are read one at a time and you can watch the progress and stop part way; whatever finished is kept. Undo restores the original extraction — including anything you had highlighted or annotated on those pages — and undoes every re-read of the document, not only the most recent one.
It is a second opinion, not a correction. The result is one model's reading of a page image, and it can be wrong in its own ways: a symbol misread, a line dropped, a table flattened. The button sits on the crop so you can see what the equation actually looks like alongside what the text made of it. If a page comes back worse, Undo it.
Everything you save — notes, bookmarks, highlights, and the Review Sheet — collects here, filterable by type. Go To Source on any item jumps back to where it came from in the document. Highlights and notes are made while reading or studying; this is where they gather for review.
Documents you open are saved on your device and listed on the start screen and in the Library tab, so you can reopen one without finding the file again. Your highlights, notes, reading position, and study history are stored on your device too, and return with the document. Reopening is instant because the parsed text is cached locally.
What Gwion sends, and when. Four things need processing that cannot happen inside a browser. If a PDF has neither a table of contents nor heading structure of its own, the file is sent so its sections can be detected — this is the only case where a whole document leaves your device, and documents Gwion can read unaided never do. Study tools send the text of the section you are working on. Quizzes and flashcards also send up to three figures from that section, as images, so questions can cover what a figure shows and not only what the text says about it. Re-reading a page sends an image of that page, one page at a time, and only for the pages you pick. Voice narration sends the text being read aloud. Nothing else leaves as an image: no figure is sent from a section you are not studying, and no page image is sent for a page you did not choose. Nothing else is transmitted: your library, highlights, notes, reading position, and study history stay in this browser.
Where that processing happens. Detecting a document's sections and generating narration run on Gwion's own servers. Study tools and page re-reading use third-party AI providers.
What not to put through Gwion. Because study tools and narration send text off your device, please don't submit sensitive personal data — anything covering someone's health, identity, or private circumstances, whether it arrives as a file or through Paste Text. Gwion is built for published and self-authored material and is not cleared for confidential records.
Study results are cached for 30 days. When a study tool runs, the result is kept on the server for up to 30 days so that asking for the same thing again is instant and costs nothing. The cache holds the generated output, not your document — the section text itself is not stored, only a fingerprint of it used to match repeat requests. Entries older than 30 days are deleted.
Rebuild clears the cached text for every document while keeping the files themselves. Use it if a document looks wrong, and it will re-parse the next time you open it.
PDF, DOCX, EPUB, ODT, TXT, Markdown, HTML, and image files are supported. RTF support is basic. Legacy .doc files should be saved as .docx first. Images and scanned PDFs use OCR when needed.
Gwion works best with structured documents, and looks for that structure in three places. If your file has a table of contents, that is used to split the text into real sections for reading and Study Mode. If it does not, Gwion reads any heading structure stored inside the file — usually present in exported documents, even when the headings are not visually larger than the body text. Failing both, documents are analyzed to recover their sections automatically. This works best on scientific papers; others are divided more coarsely.
Word pacing uses frequency data from the New General Service List and New Academic Word List (Browne, C., Culligan, B., & Phillips, J., 2013), available at newgeneralservicelist.com and used under a Creative Commons Attribution-ShareAlike 4.0 licence. Gwion groups the data into frequency bands; the values are otherwise unmodified. Development of the initial NGSL wordlist was made possible through approved access to the Cambridge English Corpus.