LORE

Ask questions about the books you own.
Every sentence cited from the text. Nothing past the chapter you're on.

Live beta · User guide · How it works · Blog · Issues

LORE is an open-source AI reader for your e-books. It turns each book into a
knowledge graph of its characters, places, groups and events, then answers
your questions from the text itself: every sentence in an answer cites the
passage it comes from, and a second pass checks each one against that passage
and cuts what the book doesn't say. Tell it how far you've read, and it never
looks past your chapter.

Features

• Answers you can check. Every sentence cites its passage; click a citation to read it in the book.
• Spoiler-safe. Set your reading progress per book. Answers, the graph, character profiles and timelines all stop at your chapter, enforced at retrieval, so the model never sees what comes later.
• Questions across a series. How a character changes from book one to book five, how a world or magic system works, who knows whom.
• Deep dives. Long, sectioned analyses with LORE's reading, the other side, and what the books leave open, all cited.
• Explore the graph. Click any character for a profile, a story timeline and their relationships, each with the passage behind it.
• Bring your own model. Gemini (the free tier works), Claude or OpenAI for reading and writing; embeddings run locally on Ollama.
• Use it from Claude. An MCP server lets claude.ai, Claude Desktop, Claude Code or any MCP client search your library and ask LORE questions.
• Private by design. Self-host it on your own machine, or use the hosted beta where each reader's library is kept separately.

Try it

The fastest way is the hosted beta at lore.jrvn.site:
request an invite on the homepage. New accounts start with A Study in Scarlet
already processed, so you can ask questions straight away. You'll need a free
Gemini API key (or a Claude or OpenAI key); the app walks you through it.

Run it yourself

Prerequisites: Python 3.12+,
uv,
Node.js 18+ and Ollama.

# 1. The local embedding model (keep Ollama running)
ollama pull nomic-embed-text

# 2. A model key: a free Gemini key from https://aistudio.google.com/apikey
export GEMINI_API_KEY=your-key-here

# 3. Install, and check everything is in place
git clone https://github.com/Jatin-Roopchandani/LORE.git
cd LORE
uv sync
uv run lore doctor

# 4. Build the web app (once)
cd web && npm install && npm run build && cd ..

# 5. Start it
uv run lore ui

Open http://127.0.0.1:8766, click + Add book, and choose an EPUB or PDF
you own. The Library panel shows processing as it goes (a novel takes a while on
Gemini's free tier); when it's done, ask away.

Prefer the command line? lore ingest, lore process and lore ask do the
same; see the user guide. Instead of the environment variable,
the key can also live in ~/.config/lore/gemini.env (mode 600).

Use it from Claude

LORE runs an MCP server, so Claude can search your library and ask LORE
questions with your reading position applied:

• Hosted beta: in claude.ai, open Settings → Connectors → Add custom connector and paste https://lore.jrvn.site/mcp.
• Self-hosted: run uv run lore mcp (stdio) or uv run lore mcp --http.

Step by step: Use your ebook library in Claude and the user guide.

How it works

book (EPUB/PDF) → chapters & passages → LLM extraction → entity resolution
               → Kuzu graph + LanceDB vectors

question → split into parts → starting entities → graph walk → passages
         → claims, each quoting a passage → check every claim → cited prose

Reading progress becomes a per-book chapter cap applied as a prefilter on every
vector search, graph query and profile, so later chapters are never retrieved.
Behaviour changes ship behind switches, each judged by blind paired evals on
held-out questions. The full pipeline, models and design decisions are in
How LORE works.

Stack: Python, FastAPI, Kuzu (graph), LanceDB (vectors), SQLite and
Postgres, Ollama (embeddings), React and TypeScript. The hosted beta runs on
Docker Compose behind a Cloudflare Tunnel (deploy/).

Project status

LORE is in beta. It works end to end on full novels and series, and the
test suite (4,000+ server tests, 800 web tests) runs on every push. Known
limits: answers are only as good as the extraction, long series take a while to
process, and a book must be one you can upload as EPUB or PDF.

Contributing

Bug reports, ideas and pull requests are welcome. See
CONTRIBUTING.md for setting up a development environment and
running the tests. To report a security issue, please follow
SECURITY.md rather than opening a public issue.

License

MIT © Jatin Roopchandani