An audiobook you can talk to.
Any book, article or PDF, read aloud, on your own device. Ask out loud, and it answers from what you have heard so far and won't spoil what's ahead, then keeps reading. It can read in your own voice.
No account, no key, no sign-up. Your books, the narration, the speech-to-text and your voice stay on your device; only the questions you ask reach the hosted brain, with the part of the book you have heard.
Made for a laptop or desktop in Chrome, Edge or Safari. Not for phones yet.
What it does
In your browser
- Any book, article or PDF. An EPUB, PDF, Word, Markdown or text file, pasted text, or a web address: drop it in, press Play, and tap any sentence to read from there.
- Ask as you listen. By voice or by typing, or hands-free from your earbuds: one press asks, two skip a sentence, three go back (Chrome and Edge). The answer is grounded at the spot you were listening to, so it won't spoil what you haven't heard yet.
- Your own voice as the narrator. Read three sentences aloud once and your books are read back in your voice; the recording never leaves your device. Try it.
- Local by design. Your books, the narration, the speech-to-text and your voice stay in your browser; only the questions you ask reach the hosted brain, and nothing is stored by us.
It also keeps bookmarks and notes by voice, quizzes you at the end of a chapter if you ask, brings your questions back later for a revisit, turns them into a study sheet and flashcards, reads formulas like a tutor, describes the figures if you let it, and keeps a long book in mind beyond the last hour: how the reader works has the details.
See it
Measured
The one claim here is that the answers stop where you are. It is measured as the effect of the position gate with the same model on both sides: 60 questions at fixed listening positions in 6 books, every answer graded blind by two judge models.
| What | Gate on (yapbook) | Gate off (same model, whole book) |
|---|---|---|
| Spoiler-free answers to questions whose answer lies ahead of you | 100% (72 of 72 verdicts) | 22% (16 of 72) |
| "The book has not covered that yet", when that was the honest answer | 96% (46 of 48 verdicts) | 8% (4 of 48) |
The run of 2026-10-22 on the brain deployed since then (its code behind a local port, the same model); the first run, against the live site on 2026-10-08, scored 72 of 72 and 48 of 48, and the two verdicts that moved are answers that give only the part already read, without a spoiler. One run per cell, the judges are language models, the questions were typed, and one of the six books is a textbook the models know by heart. What changed between the runs, the judges' prompt and every verdict are in the benchmark records on GitHub.
About
yapbook is a hobby project by Abinesh, built for listening to dense textbooks while his hands are busy. Yap all you want: any honest review is welcome, criticism, suggestions and bugs included.
There is also a version that installs on your computer and keeps everything on it; it is not published.