Read three times faster. Without losing the story. A speed reader paired with an AI companion that refuses to mention anything you haven't read yet.
"The reading companion knows where you are. It will not discuss what comes next."
It loads any EPUB or PDF, streams the text one word at a time at up to 800 sustainable words per minute, and won't say a word about what comes next.
Most reading time goes to eye movement — saccades, regressions, the half-second spent finding the next word. Rapid Serial Visual Presentation fixes the eye on a single point and pushes the text past it, so the brain keeps reading while the eyes stop hunting.
Each word is rendered with an Optimal Recognition Point pivot — the focus letter highlighted in red — and held for a calibrated delay that stretches for punctuation and shrinks for short words, rather than a flat tick.
The companion is the other half. Click any unfamiliar word for a definition shaped by the sentence around it. Hit Scene Summary and it recaps the last 150 words. Open chat and it answers anything about the story so far — then declines to predict the rest.
Generate study notes for the chapter you just finished and paste them into Notion, Bear, or Docs with formatting intact.
You can't reliably ask a model to "not spoil." So the unread chapters are made physically unreachable — the model can't leak what it was never sent.
The chat endpoint is only ever handed chapters 0 through your current position. Across 50+ books — including ones whose endings are common knowledge — it held at zero leaks.
EPUBs and PDFs flow through a single pipeline. PDF parsing is the hard half — stripping headers, footers, page numbers, footnotes, and back matter without dropping body text, with Ghostscript fallback for thin output.
Lists, tables, and figures surface in the side panel as playback reaches them. A "See Figure 2" in the body brings Figure 2 into the margin one word later — click any thumbnail for a lightbox.
They're hundreds of constructPath operators. The parser walks the operator list, clusters paths into a chart region, and crops the rasterized page — every chart in a 15-page UBS report on the first pass.
A two-pass detector — strict chapter patterns first, larger-than-body font headings as fallback for articles — plus back-matter detection so the Bibliography stops the stream where a human would.
A 167ms per-word base delay at 300 WPM, stretched for punctuation and trimmed for short words — so comprehension survives the speed instead of being sacrificed to it.
Built so the AI disappears into the background of a tool you already use — and improves it without taking it over.
100–1200 WPM with calibrated per-word delays and ORP pivot rendering. Step by word or sentence, jump back, restart.
Press C anywhere in the reader. The companion sees only what you've read. Optional spoilers toggle for after you finish.
One click recaps the last 150 words. Useful when picking up after a break.
Pause, click any word, get a definition that reflects how the sentence around it uses the word.
Per-chapter, generated on demand, copy as rich text. Lands cleanly in Notion, Bear, Docs.
Lists, tables, and figures pulled from source pages and surfaced as you reach them. Click any figure to enlarge.
50 hand-picked public-domain titles from Standard Ebooks. Drag-and-drop your own EPUBs or PDFs.
Three themes, four reading fonts, four focus-letter colors. Reading position persists per book.
One of our internal demonstrations of how AI can disappear into the background of a tool you already use — reading faster, with nothing spoiled, and the figures still in view.