What it does
Written by AI from the READMESends Zotero collections to a local Sovena server that converts attachments to Markdown with page numbers, runs OCR on scans and builds a vector index. AI clients such as Claude Desktop can then search your library over MCP.
Scanned and photocopied PDFs, common in the humanities, can't be searched or cited by page. This makes them queryable in natural language, with answers traced to a page. Needs the server and an embedding service.
What you'll need
Detected automatically- An API key for some featuresWorks without one: Local embedding and OCR models, and Zotero's local API. Needs a key (Bailian and OpenRouter): Only remote commercial embedding platforms (e.g. Bailian, OpenRouter) would need a key.
- Or: a model on your own computerIt can use a local model such as Ollama
- An older ZoteroNot working with Zotero 10
Where your data goes
We found: only contacts servers you set up. 2 addresses in 1 group.
Forks and copies
No forks listed.
The developer's response
Ratings and reviews
What people who use it say. This is separate from the grade, which is what the code does, and never changes it.
No ratings yet.