Turn what you read into what you remember.

01The problem

Studying from a book costs four apps, two rounds of retyping and a chat log you will never open again.

Reading, structuring, remembering and asking are one activity, split across tools that do not talk to each other. Every handoff loses the same thing — the thread back to the page the idea came from.

1. Step 1

   A PDF reader

   You highlight the passage that matters.

   The highlight stays locked inside the file. Nothing you highlighted is ever reviewed again.

2. Step 2

   A notes app

   You copy the passage over and write around it.

   The link back to page 148 is gone. Six weeks later you cannot prove where the claim came from.

3. Step 3

   A flashcard app

   You retype the passage a third time.

   The card drifts from both the note and the source, so most people quietly stop making cards at all.

4. Step 4

   A chat window

   You paste the paragraph in to have it translated, or explained in plainer words.

   It answers about your paste, not about your book — and the answer stays in a chat log, with no page behind it and no chance of being reviewed.

FolioNode removes the handoffs. The excerpt is the unit of work — and so is everything you get back about it, down to the translation and the answer.

03The workspace

Five surfaces, one card model underneath.

Reading, collecting, structuring, reviewing and searching are different views of the same library — not different apps you have to reconcile.

ReadStructureReviewFindCollect

A reader that stays out of the way

PDF and EPUB in the same window, with the notebook map docked beside the page whenever you want it.

• Contents, excerpts and in-page find sit in one sidebar — the same three, whether the book is a PDF or an EPUB.
• Selection rounds out to whole words, so a slightly short drag does not clip the first letter.
• Reading tone, typography and immersive mode, so long sessions stay comfortable.
• Reopen a book and you are back on the page you left, in the notebook you were working in.

04Assistance

Help that writes into your library, not into a chat log.

Three assists live inside the reader: listen to the page, translate or explain a passage where it stands, and ask the book itself. None of them is on by default, none of them is a subscription you are buying by installing the app — and what they give back can stay on the card it came from, with the page still behind it.

Listen, and still come back with excerpts

Get through a chapter while walking or cooking, pick the text back up exactly where the narration left it — and keep what struck you without stopping to type it.

• Offline by default: the voices already installed on your machine, no key and no network.
• On macOS those are the system voices themselves — including the Enhanced and Premium ones you downloaded, which a browser engine never hands over.
• Sentence by sentence, with the spoken sentence highlighted on the page and the page turning by itself.
• Press 1–5 while it reads and the sentence being spoken becomes a card in that marker colour — listening produces excerpts too.
• Start anywhere — the toolbar, “Listen from here” on a selection, or the keyboard. Better online voices are there behind a one-time notice.

Ask the book you are reading

A book usually answers your question somewhere — two hundred pages from where you thought of it. Ask searches your own copy and answers from the passages it found, which is the difference between asking a model and asking the book.

• A conversation belongs to one book, so an answer can never quietly come from a different shelf.
• Ctrl+J opens Ask beside the page, next to Map and Excerpts; it names the source it is answering about, and the list follows the book you open.
• Retrieval is hybrid: keywords catch exact terms and numbers, embeddings catch the paraphrase you actually asked for.
• Passages that were neighbours in the book are merged before the model reads them, so an argument is not handed over cut in half.
• Every answer carries the passages it used, numbered — click one and land on the page it came from. An answer with no citations is labelled unsupported rather than dressed up.

Bring your own model

A study app is not a place to hand a library to a vendor by default. FolioNode ships no model and resells no tokens — you decide where the thinking happens, and nothing is metered by us because nothing passes through us.

• One protocol — OpenAI-compatible HTTP — for every provider. Presets for Ollama and LM Studio on your own machine, for OpenRouter, and for any other compatible endpoint.
• The model list is read from your endpoint — searchable, split into chat and embedding, priced per million tokens where the provider publishes it — so you pick from what is actually there.
• Your API key is held by the Rust core, write-only from the interface, and left out of the backup archive. Each provider preset keeps its own.
• Off until you set it up. No request is made — not even to localhost — before you enable it and a test connection succeeds.
• Answers stream as they are written, and stop when you tell them to.