kindle-brain

<p align="center"> </p>

<h1 align="center">Kindle Brain</h1>

<p align="center"> Turn your Kindle highlights into a personal AI knowledge base.<br> Ask Claude questions and get answers synthesized across all your books. </p>

---

You: "What do my books say about persistence and founding companies?"

Claude reads your catalog → identifies Shoe Dog, Zero to One, Elon Musk,
The War of Art, The Almanack of Naval Ravikant → reads your actual
highlighted passages → synthesizes across 5+ books with your personal angle

You connect your Kindle highlights to Claude through an MCP server. When you ask a question, Claude follows this workflow:

1. Browse the catalog — reads CATALOG.md, a compact index of all your books (~48K chars). Each book has a one-sentence personalized description, semantic tags, and cross-book links. Claude reads all entries in one call and reasons *laterally* — a Nike biography is relevant to "entrepreneurship", a philosophy book to "leadership under pressure".

1. Identify relevant books — Claude picks 5-8 books that connect to your question, including unexpected ones that keyword search would miss.

1. Read each book — calls read_book() for each selected title. Each file contains a semantic fingerprint (what YOU highlighted most, not a generic summary), chapter summaries, and all your highlighted passages organized by chapter.

1. Synthesize — Claude weaves together insights from across your reading, grounded in the actual passages you found important.

We started with a traditional RAG pipeline — ChromaDB vector search with Gemini embeddings. It worked, but we kept hitting the same problem: vector search returns isolated snippets ranked by similarity, stripped of all structure.

When you ask "what do my books say about finding your own path?", RAG gives you 10 fragments from 10 books, ranked by cosi