Semantic search in a PDF

The tool indexes your PDF’s text into passages vectorized by a multilingual model downloaded once (~115 MB, then served from the browser cache), then ranks the passages closest to your query — even without matching words. The document never leaves your browser.

FAQ

What gets downloaded?

An embedding model (~115 MB) from HuggingFace, once, only if you enable AI search. It is then served from the browser cache.

Is the PDF sent anywhere?

No — text extraction and embedding computation run entirely on your device.

What about scanned PDFs?

Search needs a text layer: run the OCR tool first on a scanned document.

  1. Drop your PDF
  2. Enable AI search (one-time model download)
  3. Type your query and browse passages ranked by relevance