Semantic Search (Metadata RAG)

Semantic Search (Metadata RAG)

Description: Find books by meaning using LanceDB over title, authors, tags, comments, and series. Supports hybrid keyword + vector search across your entire Calibre library.

Trigger Phrases

• "Find books about [topic]"
• "Search my library for [query]"
• "What do I have on [subject]?"
• "Show me books similar to [title]"
• "Find [author] books about [theme]"

Tools

• calibre_metadata_index_build() — Build or rebuild the LanceDB metadata index. Run once per library or after large batch imports.
• calibre_metadata_search(query="...", top_k=10) — Natural-language semantic search over book metadata. Returns ranked results with relevance scores.
• rag_index_build() — Build full-text content index from ebook text (requires epub/mobi extraction).
• rag_retrieve(query="...", top_k=5) — Semantic search over full book contents, not just metadata.
• search_fulltext(query="...") — Legacy full-text search with exact phrase matching and boolean operators.
• query_books(search="...", tags=[...], authors=[...]) — Structured metadata filtering with AND/OR logic.

Workflow

1. Index check: If no index exists, call calibre_metadata_index_build() first (runs async, takes 1-5 min per 1000 books).
2. Metadata search: Use calibre_metadata_search() for broad semantic queries. Combine with top_k to control result breadth.
3. Filter refinement: Narrow results by chaining with query_books() using author, tag, or series filters.
4. Deep content search: For research-style queries, call rag_retrieve() to search inside book text. Caveat: works best on epub format.
5. Result presentation: Return title, author, relevance score, and match highlights. Include calibre book_id for follow-up actions (open, metadata edit, export).

Search Operators

• Phrase match: Use quotes in search_fulltext — "machine learning"
• Boolean: search_fulltext(query="python AND (data science OR ML)")
• Tag filter: query_books(tags=["python", "tutorial"], limit=20)
• Date range: query_books(added_since="2024-01-01")

Examples

• "Find programming books about Python and data science." → calibre_metadata_search(query="programming Python data science")
• "Search inside my books for mentions of gradient descent." → rag_retrieve(query="gradient descent", top_k=5)
• "What sci-fi books do I have about AI consciousness?" → calibre_metadata_search(query="sci-fi AI consciousness") then filter with query_books(tags=["science fiction"])