Here is a thing that happens to anyone who reads for a living. You save a paper. You highlight four paragraphs. You write a note that says “this is the argument I need for chapter three”. Two years later you remember the idea and you cannot find the paper, the highlight or the note.
DoCube is built around that specific failure. It is a local-first reading and knowledge workspace for PDF, EPUB, Markdown, web pages and video on Mac, iPhone and iPad, and its design assumes that what you read should still be findable long after you have read it. The files sit on your own devices. The annotations stay attached to the exact place in the source where you made them. And the search covers what you wrote about the document, not only what the document said.
That is the pitch. The interesting part is how much of it is actually built, and where the thing still leaves you doing manual work.
Why reading falls apart after the reading
Think about where your reading actually lives right now. A PDF in a folder. A highlight trapped inside a reader that exports nothing useful. A note in a different app that says “see the paper about X”. A video with one useful sentence at minute 34 that you will never locate again.
None of that is a storage problem. You did not lose the file. You lost the connection between the file and the reason you kept it.
DoCube’s answer is to stop separating the steps. Reading, marking, searching and reviewing happen in the same library, on the same device, against the same copy of the document. That sounds like a small thing. It is the whole thing.
A teacher who reviewed the app on the App Store described juggling hundreds of PDFs and finding one quote across all of them instantly. Another reviewer, coming from Obsidian, said the useful part was jumping straight from a note to the referenced page in a PDF. Neither of them is describing a feature list. They are describing the moment the loop closes.
What DoCube actually is
One library, several formats, one set of annotations that behave the same way across all of them. That is the short version.
The library takes PDFs, EPUBs, local videos, offline web snapshots, Markdown files, MinerU archives and scanned documents. You can bring material in from local files, by scanning a folder, through the system share sheet, by clipping a web page, or by pasting a URL on iPhone and iPad. Web pages can be cleaned up and saved as snapshots or as editable Markdown.
Organisation is folders, tags and what the app calls dynamic projects: rules over name, tag, type, import date and reading progress that generate live views, so a paper reading list or a set of course notes keeps itself current without you filing anything by hand.
One library for PDF, EPUB, Markdown and the web
Reading happens in tabs, like a browser. You can hold a paper open next to the reference it cites, or a course video next to your notes on it, and switch without losing your place.
Annotations come in several types: highlights, underlines, strikethroughs, quotations, links and Markdown notes, in six colours, across PDF, EPUB, web snapshots, MinerU documents and video subtitles. Handwriting is supported too, with multiple pens, stroke and pixel erasers, paper styles and Cornell layouts, and handwritten text is searchable afterwards.
The Markdown editor is a real workbench rather than a text box: live preview side by side, maths, Mermaid diagrams, mind maps, embedded audio and video, and a full syntax example in every new document. On Mac there is a right-click formatting menu.
The free plan, the five documents and the one-time purchase
Free gets you five documents and the whole collect, read, annotate, search and review loop. That is enough to find out whether the workflow suits you, and not enough to run a research library.
Pro is a one-time purchase, listed as the Full-platform Professional Edition at $7.99 in the US App Store. Prices vary by region, and the developer has run launch discounts, so check the current price in your own store. It removes the document limit. The developer’s product page also says it adds translation and format-conversion exports on every platform, while the App Store listing describes it only as unlocking the document limit, so treat the export features as the product page’s claim rather than a settled one.
No subscription either way. In a category where the default is a monthly fee, that is worth noticing.

Search that finds your thinking, not just your files
Most document search looks at body text and stops. DoCube searches body text, highlights, quotations, links, notes, flashcards and video subtitles in one query, and it gives priority to passages you marked. That last part matters more than it sounds: the thing you are looking for is usually something you already thought was important once.
Results can be filtered by annotation type, so you can ask for only the quotations, or only the notes you wrote yourself.
Offline semantic search with a local model
Keyword search fails when you remember the idea but not the words. Semantic search fixes that: you search for the concept in your own phrasing and the app returns material that covers it without using your terms.
DoCube does this on device. You download a local model, enable it, and the meaning-based search runs without a connection. The changelog records local embedding models arriving in version 1.5.4 to strengthen on-device semantic retrieval, and a launch listing describes the retrieval stack as BM25 plus vector search. In practice that means exact words still work, and near-misses stop being dead ends.
This is the part that separates a local-first reader from a cloud one. Your library does not need to leave the machine to become searchable by meaning.
AI summaries with source links
If you configure an AI service, it can summarise a set of search results and link each claim back to the page or timestamp it came from. You check the citation, not the prose. Summaries can be saved into the library as their own item.
You choose the provider per task: an online service such as OpenAI or OpenRouter, or a local model. The app also introduced a no-configuration assistant backed by Silicon Flow’s free API service, so the AI features can be tried without setting anything up. Usage is tracked by day and by model, which is the sort of accounting most apps leave you to guess at.
Links that land on a paragraph, not a cover
A link to a document is nearly useless. A link to page 214, third paragraph, is a different object.
DoCube stores document addresses that point at PDF pages, EPUB reading positions, Markdown paragraphs, MinerU blocks and video timestamps. Those addresses work in link annotations inside the app, in Markdown notes, and in another application entirely: copy the address, paste it into a compatible notes app, and it opens the document in DoCube at the right place.
Inside the library you can link an annotation in one document to a precise location in another. A paragraph in a paper can point at a chapter in an ebook, a clip in a lecture video, or a line in your own notes. That is cross-document linking without a graph view to maintain.
Bookmarks handle the simpler case: return to a saved page in a long PDF or EPUB when the structure is too big to hold in your head.
Scanned PDFs, video and the formats people forget
Two categories of material normally fall outside a reading app: paper that was scanned, and anything that moves.
On-device OCR for scanned documents
DoCube adds a text layer to scanned PDFs on the device, after which the text can be copied, searched and annotated like any other document. Later releases optimised memory use during OCR so it stays stable on devices with less RAM.
For complex layouts there is a second route: converting the document into a structured MinerU document, which preserves blocks and page structure well enough to link to them individually.
Video treated as a document
A video document puts the player, subtitles, chapter outline, bookmarks and annotations on one page. Embedded subtitles are read directly. Videos without them can get subtitles generated on device, on supported hardware.
Subtitles can be searched, translated and annotated, with original, translated and bilingual display modes. Every subtitle annotation carries a time range, so opening it returns to that second of the video. Add playback speed, A/B looping for shadowing, and AI-generated chapter outlines, and a lecture stops being a thing you watch once.
From highlight to review card
This is the part most annotation tools skip. Marking something is easy. Meeting it again at the right interval is the work.
Annotations become review cards automatically, driven by rules over annotation type and colour. A recent release added automatic flashcard creation from annotations, processing annotations in the same paragraph together so you do not end up with five near-identical cards from one passage.

Review uses flip cards and mask cards, with feedback of hard, okay or easy driving a spaced repetition schedule. A daily review limit can be set, and by default there is none. Weekly reading and review plans, progress tracking, an activity heatmap and review statistics keep the habit visible, which is the only reason long-term review survives contact with a busy month.
Translation, export and getting your work out
Selection translation works through the system service, with no API key required. That covers the common case: a paragraph in a foreign language you need to understand right now.
Full-text translation is the heavier feature. EPUBs, web snapshots, MinerU documents and HTML can be translated paragraph by paragraph, the job can be resumed if it stops, individual paragraphs can be retranslated, and the result can be exported as a bilingual or translation-only edition. Narrated EPUBs support synchronised audio and text, and EPUB reading itself got page-by-page navigation, two-column layouts and typography controls.
Export covers Markdown, HTML, EPUB, MinerU documents and reading-layout PDFs, depending on what the source was. Exports are independent copies, readable in other apps, which is the correct behaviour for a local-first tool: the work should outlive the tool.
Sync, privacy and what leaves the device
Changes merge through iCloud using your own Apple account. Large files can move between devices over a local peer-to-peer connection with no cloud involved, and devices on the same Wi-Fi can pair and sync the library directly.
The developer states that no personally identifiable information is collected and no private data created inside the app is stored, with only anonymous usage data gathered to improve the product. The App Store privacy label is consistent with that: it declares usage data only, not linked to your identity.
AI is the exception, and it is an honest one. When you use an AI feature, the relevant text snippets go to the provider you configured yourself. If that matters to you, use a local model for those tasks and nothing leaves the machine.
Where DoCube does not fit
Apple only. iPhone, iPad, Mac and Apple Vision. There is no Android app, no Windows build and no web client, so a mixed-device household is out of luck on the non-Apple half.
The Mac requirement is macOS 14.0 or later on a Mac with an M1 chip or newer. Intel Macs are excluded, which rules out a lot of working machines that are otherwise fine.
It is a personal library, not a shared one. Sync is iCloud and local pairing between your own devices. Nothing in the material describes real-time collaboration or a team workspace, so do not buy it expecting one.
Migration is the weak spot a reviewer named directly: moving an existing Obsidian vault in means copying and pasting, and they were waiting for batch import. Markdown folder scanning exists for importing Markdown content from a folder in one go, but a full vault migration with links intact is not described as a solved path.
The developer has not indicated which accessibility features the app supports, which is worth checking before committing if that is a requirement for you.
People Also Ask
Which document formats can DoCube open?
PDF including scanned PDFs after on-device OCR, EPUB including narrated EPUBs, Markdown, offline web snapshots, local video files, MinerU archives and HTML. Web pages can be clipped or imported by URL and saved as snapshots or editable Markdown. Office formats are not listed as supported.
Can one search cover PDFs, EPUBs, Markdown, web pages and video at the same time?
Yes. A single query returns matches from document body text, highlights, quotations, links, notes, flashcards and video subtitles, with marked passages ranked higher. Results can be filtered by annotation type, so you can narrow to quotations only or your own notes only.
Does offline semantic search work with no internet connection?
After you download and enable the local model, yes. Meaning-based search then runs on the device without a connection. AI summaries are a separate feature: with an online provider or the bundled free service they need network access, while a local model you configure keeps them on the device.
What are the system requirements?
iOS 17.0 or later on iPhone, iPadOS 17.0 or later on iPad, macOS 14.0 or later on a Mac with an M1 chip or newer, and visionOS 1.0 or later on Apple Vision. The download is 107.3 MB and the interface is available in English plus ten other languages.
The point of keeping what you read
Every reading app promises a library. Very few of them survive the question of what happens six months later, when the file is still there and the reason you saved it is gone.
DoCube’s bet is that the fix is structural: keep the files local, keep the annotation attached to its exact place, make the search cover what you wrote as well as what you read, and put the whole thing on a review schedule so it comes back. The gaps are real, particularly the Apple-only footprint, the M1 requirement and the awkwardness of importing an existing vault. But the loop it closes is the one most people never close at all.
Additional Resources
This article includes content created with AI.

