How It Works

A peek under the hood for the curious. Here's what happens when you upload a book and ask questions about it.

Book Processing

When You Upload a Book

Your book is prepared for the assistant.

Hierarchical Summaries

The book is broken into sections and sub-sections, preferring chapter boundaries where possible. Each level gets its own AI-generated summary: one for the whole book, one per section, and one per sub-section. This means no matter where you are in the book, the assistant has context at multiple levels of specificity.

Vector Embeddings

The full text is chunked into small passages. Each passage is run through an AI model that converts it into a vector embedding, a numerical representation of its meaning. This is the foundation of retrieval-augmented generation (RAG): the assistant can find relevant passages by meaning, not just keywords.

Hierarchical Summaries

Your Book

Vector Embeddings

Start
End

AI Modes

When You Ask a Question

The assistant draws from this processed data to build context for its response.

Quick Mode

Fastest

Uses your current position in the book to pull together the actual text on and around the page you're reading, plus three layers of summaries (book, section, and sub-section) and feeds them directly to the assistant. No searching, just instant awareness of where you are and what's happening around you. Best for quick clarifications and “what just happened?” questions.

Context Fed to Assistant

Book Summary
High-level overview
Section Summary
Wider context
Sub-section Summary
Narrow context
+
Local Text
Verbatim text near your position
Assistant Response

Deep Mode

Agent

Everything from Quick Mode, plus the assistant becomes an agent with tools it can call to actively search your book. It decides which tools to use, refines its searches iteratively, and builds up evidence before responding.

Agent Tools

Semantic Search

The assistant writes a natural-language query, converts it to a vector, and finds the passages with the shortest distance.

Query

“free will four great errors willing responsibility moral blame choice necessity”

Top Results

0.62

The Error of Free-Will. At present we no longer have any mercy upon the concept “free-will”: we know only too well what it is…

0.50

THE FOUR GREAT ERRORS — The error of the confusion of cause and effect. There is no more dangerous error than to confound the effect with the cause…

0.49

Everything valuable is instinct—and consequently easy, necessary, free. Exertion is an objection…

0.47

No one gives man his qualities, neither God, society, his parents, his ancestors, nor himself…

Text Search

A keyword search for when the assistant needs an exact phrase or specific term. Supports multiple terms.

free will|free-will|non-free will

Passage Retrieval

When a search result looks promising but cuts off at a chunk boundary, the assistant can pull the surrounding passages to get the full argument or find a better quote.

Web Search

Also available if the assistant needs outside context, though it rarely needs to leave the book itself.

Ask Across Library

Multi-book

Same agent capabilities as Deep Mode, but scoped across your entire library or a specific collection. Search multiple books simultaneously with results attributed back to each source. Great for cross-referencing ideas across authors or asking “which of my books discusses X?”

Navigable References

From the passages it retrieves, the assistant can quote specific lines and turn them into clickable links. Clicking one takes you straight to that spot in the reader, so you can read the surrounding context and explore the idea further.

Example Response Snippet

His central claim is genealogical: the doctrine of free will was invented to justify blame, guilt, and punishment. He says:

The doctrine of the will was invented principally for the purpose of punishment,—that is to say, with the intention of tracing guilt.p. 95 · Friedrich Wilhelm Nietzsche · The Twilight of the Idols