library-rag

Build a semantic search library by indexing PDF and EPUB documents, converting them to structured markdown, and storing bge-m3 embeddings in sqlite-vec for meaning-based retrieval. This allows you to search personal texts by concept rather than exact keyword matches.

Meaning-based retrieval uses bge-m3 embeddings to match conceptual similarity rather than exact keywords. By indexing your personal library into a sqlite-vec database, searches return results based on the underlying context and semantic meaning of the query.

Yes, an OpenRouter API key is required to generate bge-m3 embeddings for semantic search indexing. You also need the sqlite-vec extension enabled to store and query these vector embeddings within your personal library database.

Yes, you can index and query personal texts in any language. The semantic search mechanism uses bge-m3 embeddings to capture meaning across languages, enabling accurate retrieval from your library regardless of the document's original language.

PDF and EPUB files are automatically converted to structured markdown using pdfplumber and ebooklib during the indexing process. This extracts the text content into a clean format suitable for generating bge-m3 embeddings and storing them in sqlite-vec.

Limitations include the strict requirement of an OpenRouter API key for embedding generation and the need for the sqlite-vec extension. Additionally, indexing is currently limited to PDF and EPUB formats, excluding other document types from the personal library.