Noxidox - The Document AI App for iOS
Last updated on 14 September 2026
Noxidox brings AI-powered document processing to your iPhone. This page will grow with in-depth content step by step.
Overview
Noxidox turns your iPhone into a personal document archive powered by AI. You can scan receipts, contracts, or handwritten notes with your camera. Import PDF, DOCX, or plain text files. Noxidox saves the text, organizes the content, and makes it searchable. Instead of tediously naming files and sorting them into folders, simply ask a question in natural language and get the relevant section from your documents. Processing takes place on your device, so your documents stay where they belong: with you.
What is a Document AI App?
A document AI app is an application that not only photographs or manages documents, but also uses artificial intelligence to extract and categorize their content and make it searchable in natural language.
The difference from a traditional document management app lies not in how documents are managed, but in everything that happens afterward. A document management app provides an interface and leaves the rest up to you: naming, filing, and retrieving. A document AI app reads the text, recognizes what type of document it is, and extracts the details that define that document. For example, the amount and due date on an invoice, the term and notice period in a contract, or the date and issuer on a certificate. This transforms a folder of PDF files into an archive that knows its own contents. The work that, based on experience, would otherwise be left undone is thus already taken care of. The term has gained traction in recent years because it describes precisely this leap: from digitizing a document to indexing its content.
Technically, the search is powered by two complementary methods. First, Noxidox creates a semantic index for each text: An on-device language model converts sentences and paragraphs into vectors that capture their meaning, so you can find an invoice even if you search for “electricity costs” but the document only contains the term “energy provider.” Second, a classic full-text search runs in parallel, reliably matching exact terms, names, and numbers—such as an invoice number or a file number. The results of both searches are combined into a single list and evaluated together. This is where Retrieval-Augmented Generation comes into play: Before the language model formulates an answer or summary, it reads only the relevant text passages actually found in your documents and does not invent anything. Here, too, the interplay is crucial: A pure full-text search does not understand paraphrases or synonyms, while a purely semantic search quickly becomes inaccurate when dealing with exact details such as IBANs or invoice numbers. Only together do they create a search that captures both the wording and the meaning—and a response based on the data in your documents rather than on assumptions.
For you, this is most evident in the way you search your archive. Instead of trying to remember which folder contains the contractor’s invoice from last spring, you simply ask for it—and you’ll get not just the file, but the specific part of the document that answers your question. Folder structures, naming conventions, and tags thus become optional rather than mandatory. This is exactly what the term “AI” means in the Documents AI app: not just an additional feature alongside scanning, but the level at which paper is transformed into usable information. The practical difference usually becomes apparent only after a few months, once the archive has grown: a folder tree becomes more confusing with every document added, while a content-indexed archive becomes more useful with every document added.
Why on-device processing matters
On-device processing means that the analysis of your documents takes place entirely on the iPhone, not on a server in the cloud.
The most obvious benefit is data protection. The documents people scan are rarely harmless: pay stubs, doctor’s notes, lease agreements, bank statements. Anyone who uploads them to the cloud for text recognition must trust that the provider will store them only as long as necessary and reliably protect them from unauthorized access. Or, in the case of cloud-based AI systems, that the data won’t be used to train additional models. If processing remains on the device, this trust issue is eliminated because the data never leaves the device. From a GDPR perspective, this is the cleanest scenario: no data processing on behalf of a third party, no transfers to third countries, and no question of where the data is located. Data protection here isn’t a setting you can enable or forget about—it’s an inherent feature of the app’s architecture.
The second advantage is availability. Cloud-based processing is only as good as the connection it runs on. Whether you’re on a train, in the basement, on a plane, or abroad without a data plan, even the best server infrastructure won’t help. If the analysis runs locally, scanning, reading, and searching work the same way everywhere—even when the device is in airplane mode. Your archive remains fully usable. This also holds true in the long term: An archive that relies on an accessible service loses its value the moment that service is shut down.
Then there's the speed. Uploading data, waiting for a response from a service, and downloading the result takes seconds—and significantly longer for multi-page documents and with a weak connection. Modern iPhones include dedicated processing units for machine learning that handle exactly these tasks directly, without any detours.
And finally, it changes the cost structure. AI services bill based on tokens and pass that on as a subscription tier or limit. Whether you make ten or a thousand queries a month makes no difference to the app's operation.
The core features at a glance
Three features turn your PDFs, Docs, or paper documents into an archive you can work with.
RAG search
You ask a question in plain language: "How much notice does my tenancy agreement require?" — and get back the passage from your documents that answers it, along with a pointer to the source. Search works on meaning rather than character strings, so wording that differs from your question is still found.
Document tagging
Each document is categorized by document type, issuer, date, and the key metrics associated with that type. You can accept the suggested tags or customize them. Filing happens automatically, rather than being a separate task.
Summaries
For each document, the on-device AI can generate a brief summary of the key points. Whether it's a multi-page contract or a doctor's note, this lets you see at a glance what it's about without having to read the entire text again.
Document AI App versus conventional document apps
Conventional document apps solve the first half of the job: they turn paper into a tidy digital file. The comparison below shows where a Document AI App goes further — not as a verdict on individual products, but as the difference between two ways of approaching the same stack of paper.
| Aspect | Conventional document apps | Document AI App |
|---|---|---|
| Finding a document again | By file name, folder and tags you assign yourself | By asking the content questions in plain language |
| Filing and structure | You name, tag and sort every single file yourself | Category and key details are proposed as you import the file |
Who is a Document AI App for?
A Document AI App earns its place wherever paper keeps arriving that you will need again later — and nobody else is filing it for you.
- Freelancers and the self-employed
- Receipts, supplier invoices and contracts pile up all year and are needed again the moment a quarter closes. Having them read and categorised as you scan removes the sorting session that usually precedes every filing deadline, and the one receipt the tax office asks about turns up in seconds rather than in boxes.
- Households with real paperwork
- Tenancy agreements, insurance policies, repair invoices, warranties and school letters rarely live in the same place. A searchable archive on your iPhone replaces the stack of ring binders without anyone having to maintain a filing system that tends to lapse after two months anyway.
- Anyone handling sensitive documents
- If you are digitising medical letters, test results, payslips or correspondence from a solicitor, you would rather they never reached someone else's server. On-device processing makes that an easy call, because the documents never leave your iPhone and there is no consent to grant that you might regret later.
- Students and researchers
- Handwritten lecture notes, photocopied papers and seminar handouts are only worth keeping if you can find the right passage again. Full-text search across your own notes turns a stack of copies into material that still has something to offer half a semester later.
- Frequent travellers and mobile workers
- Receipts, boarding passes and expense slips appear exactly when your connection does not. Because processing works offline, the receipt is captured and usable long before it goes soft in a jacket pocket or gets left behind in a hotel room.
Common questions about Document AI Apps
These questions come up again and again when the topic turns from scanning to actually making a document's content usable — and to what on-device processing means in practice.
What sets a Document AI App apart from a scanner app with OCR?
A scanner app with OCR hands you the text of a page as a string of characters. A Document AI App goes one step further and makes sense of that text: it recognises whether it is looking at an invoice, a contract or a medical letter, pulls out the details that matter for that kind of document, and makes the page findable by its content. The difference does not show up while scanning — it shows up when you go looking for something.
Is text recognition on its own enough to make an archive searchable?
For a full-text search on exact words, yes; for anything beyond that, no. Plain text recognition has no way of knowing which figure is the invoice total and which is the VAT. Only once a language model interprets the text do recognised characters turn into answerable questions — and only then does the right passage surface even when you phrase it differently from the document.
Does it also handle handwriting, tables and pages photographed at an angle?
Yes. Text recognition on modern iPhones is not limited to neatly aligned, machine-printed originals. Handwritten notes, multi-column tables, skewed shots and poorly photocopied pages are read along with everything else. Very messy originals lower the hit rate, as they would with any recognition engine — but you get that feedback immediately, so you can simply retake the page.
Is on-device processing as accurate as a cloud service?
For everyday invoices, contracts and notes: yes. Current iPhones carry dedicated machine-learning hardware, and the models running on it are built for exactly these tasks. The gap to a server-based service is less about accuracy than about model size on very long texts — and in exchange you lose the upload, the waiting and the question of who else gets to see the page.
Do I still have to create folders and name files sensibly?
No, though you still can. Category, issuer and date are suggested as you scan, and searching works on the content rather than the filename. A folder structure becomes an option for those who are used to one, instead of a precondition for finding a document again a year from now.
What does on-device processing mean for data protection?
If the analysis happens entirely on your iPhone, nothing is transmitted to a third party — which means no data processing agreement, no transfer outside your jurisdiction and no consent to collect and administer. That matters most when you digitise records touching on health, staff or clients, where anything else would bring additional documentation duties with it.
Feature timeline
Noxidox is under continuous development. These are the most recently published product updates – taken straight from the changelog, with no editorial step in between.
Last updated:
View the full changelog