Reader — a reading room for one document

Everything Reader produces is saved against the document, so the next time you open it the work is still there:

• A summary, written once and stored.
• Answers with citations — at most five per answer, each naming a page and quoting up to 300 characters of it, so you can check the claim rather than trust it.
• Notes, flashcards, quizzes, knowledge graphs and infographics, each on its own tab.
• Exports: notes as Markdown, decks as CSV or plain text, graphs as JSON.

There is no public share link for a Reader document — the exports are how work leaves the app.

• File types — PDF, TXT, MD, LOG, DOCX, PPTX, HTML, HTM, EPUB, ODT, ODP, ODS, XLSX, CSV, TSV, RTF, SRT, VTT. Convert anything else in the Files Editor first.
• Model — the shared picker, default Auto, remembered per account.
• Chat sources — Notes, Flashcards, Quizzes, Graphs, Infographics, and web search Off / General / Scholar.
• Translate — 28 languages, with a keyboard shortcut on a selection.
• Read aloud — your browser's own speech synthesis, 0.8x to 2x. No credits, no server call.
• Deck and quiz sizes — as listed in the walkthrough above.

• 100 MB for an upload; a link Reader fetches for you is capped at 25 MB.
• 24,000 characters of document text reach the model in one prompt — the first 80% and the last 20%, with the omitted middle named in the prompt so the model knows it is not seeing everything.
• Artefacts built from your library read at most 8 documents.
• 5 citations per answer, each quoting at most 300 characters.
• A scan with no text layer extracts to nothing: its index fails and answers read the pages instead.
• Uploading a document you already have re-opens the existing one instead of making a duplicate.

Reader's own interface is English today — it does not load a translation catalogue, although it will translate the document you are reading into 28 languages.

Reader is a Starter app. Inside it, the line between free and paid is the line between reading and generating.

• Free — opening a document, page text, page images, taking notes, every export, and read-aloud.
• Costs credits — Summarize, Ask and Explain, Slides, Flashcards, Quiz, Graph and Translate.
• Costs twice — an Infographic, because it is a text brief and then an image generated from it.

You do not have to guess: Reader can price an action before you run it, so the estimate you see is the same number the ledger will record.

Can I use a scanned PDF?

Not directly. If the PDF has no text layer Reader refuses it at open and points you at the Files Editor, which can produce a searchable copy. Reader reads text; it does not do the OCR itself.

How do I know the answer is really in the document?

Every answer carries up to five citations, and each one names a page and quotes up to 300 characters of it. The quote is taken from the page, not written by the model, so you can check a claim in one glance.

What happens to a document longer than the prompt budget?

Reader sends the first 80% and the last 20% of a 24,000-character budget and names the omitted middle in the prompt, so nothing is silently dropped. For a long book, asking about a selected passage still beats asking about the whole thing.

Is read-aloud a paid feature?

No. It uses your browser's own speech synthesis, so it costs nothing and works offline.

Can I share a document and its answers with a colleague?

There is no public share link. Export what you need — notes as Markdown, decks as CSV or text, graphs as JSON — and send that.

What if I upload the same file twice?

Reader recognises it and re-opens the document you already have, with its summary, notes and decks intact.

• Open Reader — the app this guide teaches.
• A reading room for papers, with the AI on tap — the article about it.
• Scribe — write and compile the paper you are reading towards
• Files Editor — make a scanned PDF readable first
• All tutorials