Short answer
Three iPhone apps we could confirm read PDFs entirely on the device: Noema, Enclave and Off Grid AI. Each extracts the PDF's text on the phone and uses a local model to answer, so the file is never uploaded. Noema and Enclave are free for on-device use, and Off Grid AI is free with a paid Pro tier.
Several other well-known local AI apps do not do this today. Private LLM's FAQ says it does not read documents, Locally AI accepts CSV, text and code files but does not list PDF, and PocketPal AI and Apollo do not list document chat.
What “privately” has to mean
Answering a question about a PDF takes three steps: extracting the text, finding the passages that matter, and writing the answer. A cloud assistant does all three on its servers, which means uploading the file. For a contract, a medical letter or unpublished work, that is the part people want to avoid.
A private PDF chat does all three steps on the phone. Two things can quietly break that, in any app: choosing a cloud or remote model for the chat, which sends the relevant text to that provider, and turning on web search, which sends your query to a search service. Keep to an on-device model when the document is sensitive.
iPhone apps and on-device PDF chat
| App | PDFs on device | Other files | Price | Needs | Good to know |
|---|---|---|---|---|---|
| Noema | Yes | EPUB, Markdown, text, JSON, and audio or video recordings through transcription | Free | iOS 18 or later | Open source (Apache 2.0). Replies say what was read and link the passages. You choose how a document is read. |
| Enclave | Yes | Text files, images, source code | Free on device; paid Pro for cloud models | iOS 27 or later | With a cloud model selected, extracted text can be sent to it. |
| Off Grid AI | Yes | Text files | Free; paid Pro tier | iOS 17 or later | Open source (MIT). Builds a knowledge base with on-device embeddings. |
| Locally AI | Not listed | CSV, text and code files | Free | iOS 18.1 or later | LM Studio's mobile app. |
| PocketPal AI | Not listed | Not listed | Free; optional purchase | iOS 15.1 or later | Open source (MIT); vision models can read an image you attach. |
| Apollo (Liquid AI) | Not listed | Image attachments | Free | iOS 18 or later | |
| Private LLM | No | None | $4.99 once | iOS 17 or later | Its FAQ says it “does not currently support reading documents or files”. |
LLM Farm lists retrieval over documents in its README, but its repository says it is temporarily unavailable on the App Store and TestFlight, so we have left it out of the table.
How to chat with a PDF in Noema
Noema is our app. There are two ways to use a PDF in it, depending on whether you have one document or a collection.
One document: attach it to a chat
- Install Noema and download a model in Explore. Explore shows which models fit your iPhone.
- In a chat, attach the PDF from the Files app. EPUB and text files work the same way.
- Choose how Noema should read it: Automatic, Read it all, Search it, or Look up pages. Short documents are ready at once; longer ones are indexed first.
- Ask your question. The reply says what it read, and the passages are one tap away so you can check them.
A collection: build a dataset
- Install the embedding model. A chat model writes answers; the embedding model builds the searchable index. Both run on the phone.
- Import your PDFs and other files as a dataset and wait until it reports Ready. Dataset indexing explains the progress and time estimate.
- Select the dataset for a chat, or turn on Dataset Search to query everything you have indexed.
To be certain nothing leaves the phone while you work, turn on Off-Grid Mode in Settings. It blocks Noema's external requests, and local generation and retrieval keep working. The privacy and network guide lists what each feature can send, feature by feature.
What to expect, honestly
- Small models read less at once. A phone-sized model has a limited context, so for a long PDF the app retrieves the most relevant passages instead of reading every page. Specific questions (“what is the notice period?”) work better than broad ones (“summarise all 200 pages”).
- Check the passages. Retrieval can miss the right paragraph, and a model can misread one. Noema's documentation says the same: it can ask the model to cite passages but cannot promise every answer rests on the exact one, so open the source for anything that matters.
- The PDF needs real text. These apps work from the text inside the file. A scan that is only images of pages gives them little to read, so run it through text recognition first.
- Indexing takes time and battery. A large document or a big collection is slow to index on a phone. It only has to happen once per document.
- Tables and figures are harder than prose. Text extraction flattens layouts, so numbers in complex tables deserve a second look.
When a Mac helps
If you also have a Mac, it can do the heavy work without involving a cloud service. In Noema, a Mac can share an indexed dataset with your iPhone through Constellation: the original files and the index stay on the Mac, and the phone retrieves passages from it. The iPhone can also use a larger model running on the Mac. Locally AI offers a comparable link to LM Studio on a computer for models.
Frequently asked questions
Which iPhone AI apps can chat with PDFs offline?
Noema, Enclave and Off Grid AI read PDFs on the device with a local model, so they work with no connection once the model is downloaded. Private LLM does not read documents, Locally AI lists CSV, text and code files but not PDF, and PocketPal AI and Apollo do not list document chat.
Is my PDF uploaded anywhere?
Not when you use an on-device model. In Noema the text is extracted, indexed and answered on the iPhone. If you choose a remote endpoint or a cloud model in any app, the excerpts needed for the answer are sent to that provider, so pick a local model for sensitive documents. Noema's Off-Grid Mode blocks its external requests entirely.
How big a PDF can an iPhone handle?
There is no single limit. With retrieval, the whole file does not have to fit in the model's context; only the passages relevant to your question do. Larger files take longer to index, and very long documents are answered from excerpts, not from every page at once.
Can these apps read scanned PDFs?
They work best with PDFs that contain selectable text. If your PDF is a scan made of page images, add a text layer with a text-recognition (OCR) tool first, then import it.
Does PocketPal AI support PDFs?
PocketPal AI's README and App Store notes did not list document or PDF chat when we checked on November 18, 2026. It does support vision models, which can read an image you attach.
Does this work on iPad too?
Yes. Noema is one app for iPhone, iPad, Mac and Apple Vision Pro, and document chat works on each. An iPad or Mac with more memory can run a larger model, which helps with long documents.
About this guide
NoemaAI makes Noema, one of the apps described here, so read this as a maker's guide and not an independent review. What it says about Noema matches the documentation. What it says about other products comes from their own listings, sites, repositories and documentation, linked below, as they read on . These products change often. If something here is wrong or out of date, write to [email protected] and we will correct it.
Sources
- Enclave: App Store listing
- Enclave: site
- Off Grid AI: App Store listing
- Off Grid AI: repository and README
- Locally AI: App Store listing
- PocketPal AI: App Store listing
- PocketPal AI: repository and README
- Apollo: App Store listing
- Private LLM: FAQ
- Private LLM: App Store listing
- LLM Farm: repository
- LM Studio: LM Link