Indexing Your Library

Indexing is what makes your library searchable: the app reads the PDFs attached to your Zotero items, breaks them into passages, and builds embeddings it can search later.

Close Zotero first. The app reads your local zotero.sqlite database directly, and Zotero locks that file while it’s open. If indexing won’t start, see The app won’t index my library.

How it works

• Extraction — PDFs are extracted with page-level granularity, so citations can point to the exact page.
• Chunking — text is split into roughly 800-character chunks with 200-character overlap, cut on sentence boundaries.
• Metadata — each chunk carries its item ID, title, authors, year, tags, collections, item type, and page number. This is what powers both metadata filtering and citations.
• Storage — embeddings are stored in a local ChromaDB database inside the active profile.

Embedding models

Embeddings are generated on your own CPU with SentenceTransformers, downloaded automatically on first use. Choose the model in Settings:

Each model creates its own collection, so you can switch between models without re-indexing if that model’s index already exists.

Re-indexing and updates

Initial indexing builds embeddings for every PDF and can take a while for large libraries. Subsequent runs only process new or changed items. Only PDFs with selectable text can be indexed — scanned documents need OCR first.

Scope and exclusions

You can limit indexing to specific collections, and exclude Zotero collections and tags you don’t want in the index. Excluded items are never embedded — which also means they can never be retrieved or sent to a cloud provider, so exclusions double as a privacy control.