EbookRAG

EbookRAG is a small Retrieval-Augmented Generation (RAG) toolkit for EPUB libraries. It indexes local ebooks and answers questions about them by combining open-source embeddings with either LangChain or LlamaIndex pipelines.

Requirements

• Python 3.10 or newer
• uv for environment and dependency management
• Access to an OpenAI-compatible chat model (hosted or self-hosted)
• Pandoc (installed automatically through pypandoc when missing)

Setup with uv

uv venv
source .venv/bin/activate
# install core dependencies declared in pyproject.toml
uv sync

Choose one (or both) pipeline extras:

# LangChain toolchain
uv sync --extra langchain

# LlamaIndex toolchain
uv sync --extra llamaindex

If you prefer ad-hoc installation, uv pip install '.[langchain]' or uv pip install '.[llamaindex]' works as well.

Configuration

1. Copy the sample environment file and adjust the values:

   cp .env.example .env

2. Place your EPUB files inside the directory configured by EPUB_DIR (defaults to epubs/).
3. Update the variables in .env as needed:
   • OPENAI_API_KEY, OPENAI_API_BASE_URL, OPENAI_API_MODEL, OPENAI_TEMPERATURE
   • EMBED_PROVIDER and related settings to choose HuggingFace, OpenAI-compatible, or Ollama embeddings
   • Chunking parameters such as CHUNK_SIZE, CHUNK_OVERLAP, CHUNK_SEPARATORS
   • Optional QA_PROMPT_TEMPLATE or QA_PROMPT_TEMPLATE_PATH for custom prompts

Usage

• LangChain pipeline

  python rag_langchain.py

  The script builds (or updates) a FAISS vector store on first run, then drops you into an interactive question prompt. Embeddings are saved under vector_store_langchain/ with a manifest for incremental updates.

• LlamaIndex pipeline

  python rag_llamaindex.py

  This variant maintains its own FAISS index inside vector_store_llamaindex/ and offers the same interactive question loop.

Both scripts log retrieval progress and print answer references so you can trace responses back to source files.

Tips

• The toolkit supports multilingual embeddings out of the box via intfloat/multilingual-e5-base.
• If you switch to Ollama for embeddings, ensure the Ollama server is running and reachable.
• Re-run the chosen script whenever you add or update EPUB files; manifests detect changes and refresh only the affected books.