Skip to content

Chat with a PDF without uploading it

Short answer

Open the PDF in Nightjar, open the assistant, and ask. The file is read by a model on your own machine, so there is no upload step and no copy anywhere else.

Model to use
Any model reads text PDFs. Scanned files need Qwen 3.5 or Gemma 4.

Why do it locally

Most “chat with your PDF” services work by receiving your PDF. For anything confidential that is the entire problem, and it is not solved by a promise not to train on it.

How

  1. 01Open the PDF in a Nightjar tab, or drag the file into the assistant panel.
  2. 02Ask a specific question rather than “summarise this”. Retrieval picks passages by word overlap, so the words you choose matter.
  3. 03Read the citations under the answer — they point at the passages used.
  4. 04Follow up in the same conversation to go deeper on a section.

Where it falls down

  • A scanned PDF has no text layer, so it is read as images by a vision model and only the first 20 pages are processed.
  • Questions phrased in different words from the document may retrieve the wrong passages entirely.
  • On a very long file the model sees selected passages, never the whole document, so whole-file judgements are unreliable.

Other ways people ask this

  • chat with a PDF offline
  • chat with a PDF without uploading
  • How can I chat with a PDF using AI without sending whatever is in the file to a company?

Related