On-Device AI for PDF Documents

Overview

On-device AI means the model that answers your questions runs on your iPhone or iPad rather than on a server. For documents — contracts, medical letters, papers, personal records — that changes what you are willing to ask in the first place.

What runs on the device

• The AI model itself: Gemma 3 and Qwen3 are open models used for local PDF question answering.
• Document preparation: OCR, indexing and retrieval of relevant content.
• Inference: the answer is generated on the device.
• Citations: the page an answer came from is tracked so it can be shown.

The document pipeline

A question about a PDF goes through several steps before an answer comes back. Understanding the order makes the privacy claim concrete rather than abstract.

1. The PDF is opened on the device.
2. If the document is scanned, its pages are OCR-recognised so the text exists as text.
3. The document content is indexed locally.
4. The relevant content is retrieved for the question.
5. The local AI model generates an answer.
6. The answer is shown with a page citation.

What the network is used for

According to the product description, the network is used for one thing only: downloading AI model weights. The recommended on-device model is about 1 GB and is downloaded once. Your documents are not uploaded.

Offline after download, precisely stated

The accurate description is that AI document chat works offline after the AI model has been downloaded. This is not the same as claiming the app is offline out of the box — the model download happens first.

Why on-device matters for documents

Uploading a document to an online service is a decision about trust. On-device AI removes that decision from the workflow: the document is processed where it already is, and the answer arrives without the file leaving the device.

Key facts

• Open models: Gemma 3 and Qwen3
• Model downloaded once (about 1 GB)
• AI chat works offline after download
• Documents are not uploaded
• Answers include page citations

Related privacy detail

For how on-device processing is described from a privacy standpoint — including what the network is used for and how the App Store privacy label is attributed — see On-Device AI Privacy and PDF JustFit Privacy.