This works with any document that supports highlighting in DoCube. This guide uses yellow highlights in a PDF on an iPad as an example, from setting up a rule to completing a review.

Before you start​

Choose a document and check that you can select and highlight its text in DoCube.

Start with one concept or a short passage. Try the full workflow before applying it to more reading material.

Five steps to review your highlights​

1. Open the document you want to read​

Open the document in your DoCube library. If you are using a local file that you have not imported yet, open the import menu, choose the local file option, and select your file.

2. Set up an annotation rule​

Open DoCube settings and select the annotation section, marked with a pencil icon. In Auto Flash Card, tap the plus button.

Choose Highlight, select yellow, and enter a category such as Key concepts. New annotations that use this type and color will be added to review.

To hide the answer during review, enable the crossed-out eye button on the left of the rule. This only changes the review view; the highlighted text stays visible during normal reading.

Illustrations recreate DoCube’s iPad interface using example reading material.

3. Highlight with the matching color​

Return to your document, press and hold to select the text you want to remember, and open the annotation panel. Select yellow, then tap [Key concepts]. The label comes from the category you entered in the previous step.

Matching annotations are added to review automatically. You do not need to copy the passage into another card.

Select one concept or a short passage at a time. Leave enough surrounding text to help you recall the hidden part; hiding a whole paragraph can make the task unclear.

4. When a review is due, recall the text and reveal it​

Return to the DoCube home page and open a due plan in the review area. A newly created card may not appear in today's queue immediately. Follow the schedule shown in the app.

With concealment enabled, you see the source passage with the selected text covered and the surrounding content still visible. Try to recall the hidden text, then tap the covered area to check it.

5. Rate how well you remembered it​

After revealing the text, check your answer and choose H, M, or E for hard, medium, or easy. DoCube uses your choice to schedule future reviews. The buttons show the applicable intervals.

If you could not remember the passage, choose accordingly. The rating tells DoCube how often you need to return to it.

Common questions​

Do I need to configure an AI service?​

No. This workflow uses annotation rules in Auto Flash Card to add your selected source text to review. It does not require AI to write questions or answers.

Why is my new highlight missing from today's review?​

Check that its annotation type and color match the rule, and that you have entered a card category. Matching cards follow a review schedule; creating a card does not necessarily make it due immediately.

For your first attempt, set up the rule before creating a new highlight, then follow the steps above.

Does this work with scanned PDFs?​

Run OCR first and check that you can select text. Then set up the rule and annotate as described above. Check the recognized text before highlighting, especially formulas, symbols, and technical terms.

Can I create cards manually instead?​

Yes. After selecting text, you can also create a flip card or mask card from the card menu. Those let you compose card content yourself. The annotation rule in this guide keeps the original context for reviewing passages you highlighted while reading.

Try it with one document and one concept. Download DoCube from the App Store, or see the review demo on the home page.

After the initial model download, recognition works offline and does not require uploading your video to a speech recognition service. This guide walks you through downloading the model, importing a video, generating subtitles and checking the result.

Before you start​

If you have not installed DoCube yet, download it from the App Store. Then make sure you have:

• An Apple silicon Mac (M1 or newer) running macOS 14 or later.
• An internet connection for the initial model download: roughly 1 GB, depending on the model recommended in Settings. Recognition works offline afterward.
• A video that plays correctly and has an audible soundtrack.

Generate subtitles with Whisper in four steps​

1. Download the Whisper subtitle model​

In DoCube, press ⌘, to open Settings. Select AI Assistant, find Subtitles in the Local Model section, and click Download on the right of that row. DoCube recommends a model for your computer, so you do not need to choose one yourself. Wait for the download to finish before continuing.

The illustrations are drawn from DoCube's interface. The video and subtitle content are examples.

2. Import your video​

Click Import / New at the top left of the main window, choose Local file, and select the video to import. Common formats such as MP4 and MOV are supported.

If the video has a readable embedded subtitle track, DoCube uses it first. You can go straight to step 4 to check the subtitles.

3. Select Generate Subtitles​

Open the imported video and select Generate Subtitles in the toolbar above it. Wait for processing to finish; longer videos take more time. If a task stops after some progress has been saved, reopen the video and select Continue Generating Subtitles to resume it.

For your first attempt, try a video around ten minutes long to check the results.

The button shows Generating while recognition is running. When DoCube notifies you that subtitle generation is complete, you can check the result.

4. Check the subtitles​

If the panel on the right is closed, click the sidebar button at the far right of the toolbar above the video. Then select the subtitle speech-bubble icon at the top of the panel to open the subtitle list.

The list highlights the current sentence as the video plays. Select a sentence to jump to its time in the video.

Automatic subtitles can contain mistakes, particularly in names, technical terms and numbers. Right-click a sentence in the subtitle list and choose Edit to correct it. Subtitles with existing annotations cannot be edited, so check the text before annotating it.

These subtitles play alongside the video in DoCube. They are not burned into the original video image.

What you can do next​

• Watch with bilingual subtitles. If you need another language, configure an AI service and translate the subtitles. You can then switch between the original, translation and bilingual display. When you use an online service, the relevant subtitle text is sent to that provider.
• Listen to a sentence again. Use the current-subtitle loop to repeat a sentence, or A-B looping for a longer passage.
• Find something you heard. Search the subtitle text and return to the matching moment. See Search video transcripts offline on Mac.
• Keep a timestamp in your notes. See the guide to source-location links.

Frequently asked questions​

Do I need to install Whisper separately to use DoCube?​

No. DoCube includes Whisper speech recognition. Download the subtitle model as described in step 1; you do not need to install Python or set up a command-line environment. Generating original-language subtitles does not require configuring an online AI service either.

Does subtitle generation need an internet connection?​

The initial model download needs internet access. Once the model is fully downloaded, recognition runs on your Mac and works offline. Downloading a model again or updating it still requires a connection.

Is my video uploaded?​

Local subtitle generation does not upload your video. Features such as translation and AI chapters pass relevant subtitle text to your configured AI service. Check the provider's data-handling rules before using an online service.

Does it support Chinese videos?​

Yes. DoCube detects the spoken language automatically. Check mixed-language speech and technical terminology after generation.

What if my MKV video will not open?​

DoCube does not currently accept MKV, WebM or AVI. Convert the video to a compatible MP4 file with a video conversion tool before importing it. Renaming the file extension does not convert the video.

Do I get subtitles or a new video with subtitles burned in?​

These four steps create original-language subtitles that you can display, check and use to jump between sentences in DoCube. They do not export an MP4 with subtitles burned into the picture.

If you also need a separate subtitle file, complete a subtitle translation first. You can then export translated or bilingual VTT files from the translation panel. A VTT file stores subtitle text and timing information for use with a compatible player.

Check search with a passage you can see​

Open a PDF with a text layer and search for a short phrase visible on its page. Confirm that a result returns to that page, then try the same term in the library search. This separates missing text from incomplete library indexing.

Next, search the library, open a result and check the source document and matching location. For a scan, run OCR first and repeat both checks.

Start with a 10-second test: text PDF or image scan?​

Open one PDF and try to select a sentence:

• If individual words and lines can be selected, the file already has a text layer.
• If only the whole page can be selected, or no text selection is possible, it is probably an image-only scan.
• If selection works but copied text is garbled, the layer exists but its encoding or OCR quality is poor.

This distinction matters because search software can only index text it can actually extract. A page that visibly contains words does not necessarily contain searchable characters.

Pick the search method that fits the collection​

Spotlight is a useful first attempt on macOS. Apple says its regularly updated, private index stays on the device, although indexing a large collection can take time. See Apple's explanation of Spotlight indexing.

For a folder of PDFs, Acrobat's Advanced Search can search a selected location instead of only the current document. Adobe also supports catalog indexes for large, stable collections. See Adobe's guide to searching PDFs.

Folder search becomes less useful when the same project also contains EPUBs, saved webpages, Markdown notes, video transcripts, highlights, and comments. A local document library can index those sources together and preserve a route back to the matching page, chapter, or timestamp.

Why scanned PDFs need OCR first​

An image-only PDF is a stack of page pictures. Optical character recognition turns visible characters into machine-readable text that selection and search tools can use.

Three habits make OCR safer:

1. Choose the correct primary language, adding a secondary language for mixed-language documents.
2. Keep the original file or a backup before processing a large batch.
3. Sample-check names, numbers, formulas, and multi-column reading order.

Low-resolution pages, formulas, code, and complex columns are common failure points. OCR makes a scan searchable; it does not make the recognized text authoritative. Adobe likewise describes OCR as the step that creates selectable, searchable text from page images and recommends reviewing the result. See Adobe's scanned-document OCR guide.

Build a searchable local PDF library in DoCube​

For a collection you expect to keep using, the workflow is:

1. Import files from the Add menu, or use directory import to scan a folder in bulk. Enable “Keep Directory Structure” when the source hierarchy matters.
2. Open an image-only PDF and choose Text Recognition from the Mac toolbar. Select the primary and any secondary languages, then start OCR.
3. When recognition finishes, DoCube reopens a version with a text layer. The imported original PDF is not directly rewritten.
4. Allow newly imported or recognized text to finish parsing and indexing. A large collection is processed in stages.
5. Select “Search in DoCube” in the tab bar, enter a keyword, exact passage, or short question, then press Return.
6. Filter results to PDFs, a tag, a source document, or a content source such as body text or annotations.
7. Inspect the result snippet, then open the source. A PDF result returns to the matching page and location.

DoCube's library search includes PDF, EPUB, HTML, Markdown, MinerU documents, and video subtitles. It also indexes highlights, quotes, links, Markdown annotations, and cards, so a phrase you wrote yourself can lead back to the original source.

When semantic search helps​

Keyword search is best for names, identifiers, technical terms, and remembered phrases. Semantic search becomes useful when:

• You remember a concept but not the author's wording.
• Several authors use different terms for the same idea.
• You want related passages rather than exact strings.

In DoCube, you can download the local QWen3 Embedding model and enable Semantic Search for a new query. The results merge BM25 keyword ranking with vector matches. After the model is downloaded, vector indexing and retrieval can run offline on the device. The first pass over a large library still takes storage, memory, and time.

Semantic similarity is not factual agreement. Always open the source when numbers, conditions, or conclusions matter.

If a known phrase does not appear, troubleshoot in this order​

1. Search inside the current PDF. If that fails too, inspect the text layer or OCR result.
2. Shorten the query. Start with two or three consecutive words instead of a long question.
3. Wait for indexing. Newly imported, recognized, or edited content may not be searchable immediately.
4. Remove filters. Check document type, tag, source, and case-sensitive options.
5. Check permissions and encoding. Encryption, corruption, and unusual font maps can block extraction.
6. Rebuild the index last. Reindexing cannot repair a missing text layer.

Local search is not the same as AI summarization​

Keyword indexing, OCR, and local embeddings can stay on the device. AI summaries of search results are a separate operation: when enabled, the query and selected result snippets are sent to the AI provider you configured. For confidential papers, contracts, or internal records, treat retrieval and online summarization as two different privacy decisions.

Frequently asked questions​

Can Spotlight search every scanned PDF?​

Not reliably. Spotlight depends on indexable content. Image-only pages, damaged text layers, or files that have not finished indexing can be absent. OCR the file first, then test a distinctive phrase.

Does OCR change the visible PDF pages?​

It depends on the tool. DoCube creates a recognized version while preserving the page appearance and does not directly rewrite the imported original. Before using another batch tool, check its output and overwrite policy.

For thousands of PDFs, should I use keyword or semantic search?​

Build a dependable keyword index first. It is explainable and remains best for names, numbers, and exact terminology. Add semantic retrieval as a second route rather than replacing keyword search.

If you are also deciding where annotations and reading notes should live, continue with How to Manage PDF and EPUB Notes Locally. DoCube is free to try on the App Store, so you can validate the workflow with a small library first.

Check a representative chapter first​

For the first run, choose a short chapter or small EPUB containing headings, an image and ordinary paragraphs. Open the translated book and check its contents links, paragraph order, image placement and resizable text before processing a larger book.

If your material includes equations, code, footnotes or tables, compare one example of each against the original. Check both structure and meaning, rather than relying only on a completed task status. Keep the original book and record the service and model used so you can repeat the job if needed.

Why copy-and-paste translation breaks ebooks​

An EPUB is not one long plain-text file. The W3C EPUB 3.3 specification defines a single-file container that can include XHTML, CSS, images, fonts, navigation, metadata, and other publication resources. Its table of contents and default reading order are structured too.

That is why these approaches commonly lose information:

• Copying a whole chapter into a generic text box and keeping only the returned translation.
• Converting the book to Word or plain text, then trying to rebuild the EPUB.
• Splitting text by visible line breaks without respecting headings, lists, footnotes, and links.
• Converting to PDF in the name of “preserving layout,” which removes the EPUB's reflowable reading behavior.

What must survive is not a pixel position from one screen. It is the publication structure: which paragraph belongs to which chapter, where images appear, which spans are headings or emphasis, and how navigation reaches the content.

Choose the method by what you need afterward​

If one sentence is blocking you, selection translation is faster. Full-document translation becomes worthwhile when you want to read continuously, compare source and translation, or export a complete book.

Check four things before translating​

1. Make sure the file is DRM-free. Store-protected ebooks generally cannot be processed by third-party tools.
2. Inspect the table of contents and a few chapters. Translation will not repair images or navigation that are already broken in the source.
3. Identify reflowable versus fixed layout. Novels and most nonfiction are reflowable; comics, art books, and complex textbooks may use fixed layout and are more prone to overflow.
4. Keep the original EPUB. A translation should be exported as a new file, not overwrite the only source copy.

Also make sure you have the right to translate and export the material. Personal reading, public distribution, and commercial republication are different permission scopes.

Translate a complete EPUB in DoCube​

DoCube translates EPUB body content paragraph by paragraph and keeps a correspondence between source and translated passages. The practical workflow is:

1. Import and open the EPUB. Confirm that navigation, images, and body text work before starting.
2. In Settings → AI Assistant, configure an online provider and model for document translation. Full translation sends the required text to the provider you select.
3. On Mac, choose Full Translation in the top toolbar. On iPhone or iPad, open it from the reading actions menu.
4. Choose a target language and start. You can leave and continue unfinished work later; completed language versions can be reopened directly.
5. Read in side-by-side, stacked, or translated-overlay mode, and swap source and translation positions when useful.
6. If one paragraph fails or reads poorly, retranslate only that paragraph. The previous text is replaced only after a new result succeeds.
7. From the translation panel, export either a translation-only EPUB or a bilingual EPUB that keeps source and translation together.

Translated-file export is a Pro feature. It creates a new file; the original EPUB remains in the library.

What “preserved formatting” should mean in testing​

No automatic pipeline can promise pixel-perfect output for every EPUB. Check at least:

• The table of contents still opens the correct chapters.
• Chapter order is complete, with no duplicated or missing sections.
• Headings, lists, quotations, footnotes, and links retain their semantics.
• Images remain present and captions stay near the intended image.
• Emphasis and paragraph boundaries have not been split incorrectly.
• Longer translated text does not overflow tables, poetry, controls, or fixed layouts.
• Names, terminology, numbers, negation, and units are accurate.

Open the export in two reading systems and at more than one window width. A reflowable book paginates differently by device, so looking correct in one viewport is not enough.

Bilingual EPUB or translation-only EPUB?​

Choose based on the next reading task:

• Bilingual output is better for language learning, professional material, and frequent source verification. It is also longer.
• Translation-only output is calmer for continuous reading, but errors require returning to the source book.
• In-app comparison without exporting yet is best during proofreading. Export a final copy after the translation is stable.

For research papers, contracts, technical standards, and medical material, keep the source available. Machine translation should not replace verification of consequential wording.

EPUBs most likely to need manual repair​

Pay extra attention to:

• Fixed-layout ebooks, comics, and illustrated books.
• Mathematics, code, vertical writing, and ruby annotations.
• Heavy use of footnotes, cross-references, or popup notes.
• Sentences split across many inline formatting elements.
• Books that depend on scripts, external assets, or custom fonts.
• Source files that are already non-conforming EPUBs.

For these books, translate and export one chapter first. Verify navigation, images, and paragraph structure before spending time or API cost on the rest.

Privacy, cost, and terminology consistency​

The EPUB can remain stored and read locally, but DoCube's current full-document translation uses an online AI provider configured by the user. Relevant text is transmitted to that provider and may incur usage charges. Review its data policy before translating sensitive material.

Long books introduce a second problem: terminology drift. A model may translate the same concept differently in later chapters. Define important terms in advance when possible, then sample-check headings, names, and repeated domain vocabulary. High-stakes output still needs translation memory or human review.

Frequently asked questions​

Can DoCube translate a whole EPUB entirely offline?​

Not currently. Full-document translation requires a user-configured online model. Whether system selection translation can work offline depends on installed operating-system language resources, and it is not a full-book export workflow.

Can I still change the font size after translation?​

A reflowable EPUB should retain ebook reflow, with font size and content width controlled by the reading system. Fixed-layout publications may not offer the same flexibility.

Does bilingual export overwrite the original book?​

No. DoCube exports a new bilingual or translation-only EPUB from the translation panel, while the imported source remains in the library.

If you want the source book, translated copy, and reading notes to remain organized locally, continue with How to Manage PDF and EPUB Notes Locally.

Confirm that a transcript actually exists​

Open the video's subtitle list and check for readable cues. The capture below is a real empty state from a dedicated test video: playable media does not by itself mean that a transcript is available for search.

Once cues are available, search the current video for a word visible in the list, open a match and check its time and audio. Then test library search. If a transcription job finishes but the list remains empty, investigate the transcript before treating the index as ready.

Video search is really three connected tasks​

1. Transcription: recognize speech from the audio track.
2. Alignment: store a start and end time for each subtitle segment.
3. Indexing: add the text to search while preserving its timestamp.

A plain TXT transcript can be searched, but it is awkward to verify against the source. Timestamped subtitles without a library index still make you open videos one by one. A complete workflow solves all three problems.

OpenAI Whisper is a general-purpose speech-recognition model that supports multilingual recognition, language identification, and speech translation tasks. A local application such as DoCube can run recognition on the Mac after the model files have been downloaded, without uploading the audio to a transcription website.

Choose the approach by your actual need​

That last distinction is important. Subtitle search finds spoken language. Slide text, code on screen, and titles that nobody reads aloud do not become searchable merely because a Whisper transcript exists.

Generate and search local video subtitles in DoCube​

1. Prepare the local model​

Open Settings → AI Assistant and download a Whisper model from the Local Models section, then select it for video subtitles. The initial model download requires a network connection. Once the files are ready, subtitle generation itself can run offline.

Larger models generally require more memory and processing time. Before a long job, leave enough storage for the model, temporary audio, and subtitle data, and close other memory-heavy applications.

2. Import the video and create subtitles​

If this is your first time, follow the illustrated guide to generating video subtitles with Whisper on Mac to download the model, import your video and check its subtitles before returning here to search.

Import a local video into DoCube and open the video document:

• If the file contains a usable embedded subtitle track, you can use it directly.
• If it has no subtitles, choose Generate Subtitles to start local recognition.
• An interrupted or incomplete job can continue later.
• Keep the app and current document open while recognition is active and watch the progress indicator.

When subtitles are ready, the subtitle list follows playback and highlights the current cue. Selecting a cue jumps to its start time.

3. Search the current video​

Open Find in Document and enter a word or phrase. This searches the current subtitle text; it does not re-analyze the audio or frames. Results include the matching cue and context, and selecting one jumps to that time.

This is fastest for names, terminology, and remembered quotations. If subtitles are incomplete or the recognizer misspelled the word, the match will be absent.

4. Search across a video library​

Video subtitles also enter DoCube's library-wide index. Select “Search in DoCube” in the tab bar to search multiple videos together with PDFs, EPUBs, saved webpages, Markdown, and your annotations.

Library results retain subtitle timestamps. Opening one returns to the matching moment instead of merely opening the beginning of the video.

5. Turn a useful moment into retrievable knowledge​

Select subtitle text to create a highlight, quote, Markdown annotation, link, or card. The annotation stores both the selected words and the time range, so opening it later still returns to the source moment.

To cite a segment from another note, copy the DoCube address at that video position. The address identifies the video and its playback time.

What offline transcription actually protects​

Local Whisper keeps the audio from being uploaded to a third-party transcription website. That is useful for meetings, interviews, course recordings, and unpublished material.

Later features can have different boundaries:

• Local subtitle generation and subtitle search can run offline after model download.
• Subtitle translation sends relevant text to the online provider you configure.
• AI-generated chapter outlines also depend on the configured online service.

Do not assume that every downstream operation is local merely because the transcript was created locally.

Improve transcription and search accuracy​

• Prefer a clear track with close speech, limited music, and little room echo.
• Choose a model that fits the Mac's memory and the recording length.
• Review and correct names, acronyms, numbers, and technical terms.
• If a query fails, try a shorter phrase, stem, or likely phonetic spelling.
• For multilingual recordings, confirm that the detected source language is sensible.
• Return to the audio and frame for important claims; do not treat the transcript as the source of record.

When subtitle search is not enough​

Additional visual processing is required when you need to:

• Find text on a slide that the speaker never said.
• Locate a product, person, scene, or action.
• Search menus or code in a silent screen recording.
• Retrieve by visual similarity rather than spoken language.

Those tasks need keyframe extraction, OCR, image captioning, or visual embeddings. They complement speech transcription; they are not the same Whisper capability.

Frequently asked questions​

Can the first run be completely offline?​

No. The local Whisper model must be downloaded first. After a complete download, subtitle generation and transcript search in DoCube can work offline.

Can a result jump directly to the matching video time?​

Yes. Both in-document subtitle results and library-wide video results preserve time locations. Subtitle annotations preserve their cue range too.

Does Whisper search text visible in video frames?​

No. Whisper processes the audio track. On-screen text needs OCR or a visual index, and unspoken content will not appear in transcript search.

Can an automatic transcript be used as an official record?​

It should be reviewed first. Names, numbers, negation, and domain terminology are common error points. Formal minutes, quotations, and legal material should be checked against the original audio.

To connect a useful video moment with your reading notes, continue with How to Link Notes to Exact Places in PDFs and EPUBs.

Check a complete round trip with your own files​

Choose a PDF, a webpage and a short video, then write a note about one question. Before importing a large collection, check three things:

1. Reopen the note and confirm that its text and inserted material remain readable.
2. Follow a source action back to the page, passage or video range it refers to.
3. Search a distinctive term and verify that the note or source content is actually found.

Successful sync is different from having a backup. Export important material to another location, check that the files reopen, then plan continuing backups. The connected-notes workflow shows how these pieces fit together.

Separate the five kinds of data​

A long-lived reading library contains at least:

1. Source documents: PDFs, EPUBs, and their image or font resources.
2. Reading state: current page or chapter position, bookmarks, and progress.
3. Source annotations: highlights, underlines, quotations, and short comments tied to a location.
4. Independent notes: Markdown or structured notes that can combine ideas across documents.
5. Search indexes: derived data built from documents and annotations; disposable and rebuildable, never the only copy.

Writing everything back into the PDF or EPUB appears simple but is limited by each format and reader's compatibility. Separating everything without links loses the route from a note back to evidence. A practical local-first design keeps source files independent, stores structured annotations in the library, and records enough location data to reconnect them.

PDF and EPUB locations are fundamentally different​

PDF pages are fixed, so a location can usually be expressed as a page plus an area or selection. EPUB content reflows when font size, viewport, or device changes, so “page 83” is not a stable address. A chapter and content position is more durable.

That difference should shape the note system:

• A PDF note should return to a specific page and passage.
• An EPUB note should follow the content location, not a page number produced by one layout.
• Replacing or heavily editing the source can invalidate either locator.
• A screenshot preserves appearance but discards searchable text and portable structure.

A tool that manages both formats must respect these location models instead of treating every document as a generic attachment.

Three common ways to store reading notes​

These methods are complementary. Keep short reactions near the source and longer thinking in independent notes, then connect them with stable links.

A sustainable local workflow​

1. Establish a document entry point before designing a taxonomy​

Bring existing PDFs and EPUBs into one library. A first migration can preserve the source folder hierarchy, while a small set of tags represents themes that cross folders.

Folders work for stable ownership such as a course, client, or project. Tags work for states and overlapping subjects such as “to read,” “methodology,” or “citation candidate.” Avoid building dozens of nested levels before search and links have proven useful.

2. Capture the smallest sufficient context while reading​

A useful annotation should know:

• Which document it came from.
• The exact source text.
• Its PDF page, EPUB content position, or video time range.
• Whether it is a highlight, quote, link, long note, or card.
• When it was created or edited.

Do not force every thought into a tiny margin comment. A reaction to one sentence can stay attached to the source. Analysis that compares several sources belongs in a separate Markdown note with links back to evidence.

3. Put annotations in the search index​

As the collection grows, you often remember something you wrote rather than the filename. Full-text search should cover body text, highlights, quotations, links, and notes while preserving their sources.

DoCube's library search covers PDF, EPUB, HTML, Markdown, MinerU, video subtitles, and multiple annotation types. Results return to pages, chapters, content blocks, or timestamps rather than merely opening a document cover.

4. Connect long notes to sources with location links​

Copy a docube:// address at the current reading position and paste it into Markdown or another link-capable notes app. The long note remains editable text while one click returns to the evidence.

The address does not contain the source file. When moving notes to another device, make sure the target document has been imported or synchronized there too.

5. Treat sync and backup as different systems​

Local-first does not mean keeping only one copy:

• The local library is the primary working copy.
• iCloud can synchronize changes among Mac, iPhone, and iPad.
• Sync propagates edits and deletions, so it does not replace an independent backup.
• Export important source files, translations, and notes periodically, and retain recoverable history.

Before deleting a document, check whether annotations, cards, or other notes still refer to it.

How DoCube stores the workflow​

DoCube copies documents into an on-device library. Its PDF annotations are stored separately instead of being written back into the original file. PDFs, EPUBs, webpages, Markdown, MinerU documents, and videos can be organized, searched, and linked together.

A useful small-scale test is:

1. Import one folder and verify titles, covers, and extracted text.
2. Run local OCR on any scanned PDFs.
3. Create a few highlights, quotations, and Markdown annotations in a PDF and an EPUB.
4. Create one independent Markdown note and paste several exact-location links into it.
5. Search for words you wrote and confirm that the results return to the source.
6. Before enabling iCloud sync, confirm that local storage and export behavior match your expectations.

The free edition supports up to five documents, enough to validate the complete loop with real material before moving a larger library.

Checklist for a local reading-notes tool​

• Does it read PDF and EPUB directly rather than treating them only as attachments?
• Can it OCR scanned PDFs locally?
• Are annotations searchable, with a route back to the exact source?
• Do EPUB links survive font-size and viewport changes?
• Are long-form notes open or exportable?
• Can it export source files, translated versions, and independent copies?
• Is synchronization optional, and where does primary data live?
• What associated data is deleted when a document is removed?
• Can the search index be rebuilt from source documents and notes?

Frequently asked questions​

Should annotations be written directly into the PDF?​

That makes sense when a marked-up PDF must travel to other readers. For unified search, links, and review across PDF, EPUB, and video, structured library annotations are more flexible, provided there is a clear export and backup path.

Why should an EPUB note not store only a page number?​

Most EPUBs repaginate when font size, window width, or device changes. A page number is the result of one layout, not a stable content address. Store the chapter and content position instead.

Does local-first mean iCloud cannot be used?​

No. Local-first means a complete usable primary copy exists on the device and core reading does not depend on continuous access to a service account. iCloud can be a user-controlled synchronization channel, but sync is not backup.

What matters most when moving to another tool?​

Keep the original PDFs and EPUBs, readable note text, annotated source text with locations, and required image assets. An app-only database with no readable export raises long-term switching costs.

Next, read How to Link Notes to Exact Places in PDFs and EPUBs to turn “stored locally” into “able to return to the source later.”

A real link between two documents​

This device capture shows a PDF location URL being entered into a link annotation in an EPUB. The URL contains both a document identifier and a location component: it is meant to return to a saved place, rather than merely open a file.

To check your own link, copy a source location, paste it into another note or link annotation, save, leave the source page and follow the link back. Check where it lands, not only whether the URL becomes clickable.

For a workflow combining links, excerpts and video clips, see bringing evidence from different sources into one note.

Document links and location links solve different problems​

Imagine a key conclusion inside a 400-page report:

• A document link reopens the report.
• A page link opens the relevant page.
• A location link returns close to the passage or selection you were reading.

The difference feels small with ten files. Months later, when all you remember is “that report mentioned it,” reopening the cover is almost the same as starting another search. The useful link completes the last mile from note to evidence.

PDF annotations and linked notes are not the same thing​

A practical division is to keep short reactions on the source and long analysis in independent notes, with exact source links beside every important claim.

Store locations according to the format​

• PDF: document identity, page number, and a more precise page position when available.
• EPUB: document identity, chapter, and content position rather than a page number that changes with font size.
• Markdown: document identity plus a heading or custom anchor.
• Saved webpage: the saved document identity and paragraph position, not only a live URL that may change.
• Video: document identity and playback time, ideally with the subtitle range too.

No locator is indestructible. Replacing a file, heavily editing chapters, deleting headings, or regenerating content can make old links inaccurate, so sample-check important references after structural changes.

Copy and reuse an exact location in DoCube​

DoCube uses docube:// addresses to represent documents and reading positions. The basic workflow is:

1. Open the target PDF, EPUB, Markdown, MinerU document, saved webpage, or video.
2. Navigate to the page, passage, block, or timestamp you want to cite.
3. Use the document-address action to copy the current DoCube location.
4. Paste the address into a Markdown link, a link annotation, or another notes app.
5. Open it immediately and verify the target.

In Markdown, wrap the raw address in a readable title:

[Original argument about retrieval ranking](docube://...)

Changing the text inside brackets does not change the target. Editing the URL does.

Link one source passage to another document​

To make a passage in a PDF or EPUB point directly to other evidence:

1. Open the target and copy its current location address.
2. Return to the source document and select the words that should carry the link.
3. Choose Link from the selection actions, paste the target address, and save.
4. Open the new link annotation and verify the destination.

This is useful when you want to:

• Connect a paper's conclusion to its method or data page.
• Link a textbook passage to a definition in another chapter.
• Compare two translations of the same work.
• Jump from criticism to the exact passage under discussion.
• Connect course notes to a precise moment in a lecture video.

A link annotation is a one-way relationship. It does not automatically create a “linked from” list in the target document. Tab history can take you back through a visit, but history is not a backlink either.

Why independent Markdown works well as the connection layer​

An independent note must stay easy to reorganize while retaining a route to original material. Markdown fits that layer because the body remains readable text and links live directly in the content.

One topic note can point to:

• A page inside a PDF.
• A passage in an EPUB.
• A paragraph in a saved webpage.
• A timestamp in a video.
• A heading or anchor in another Markdown note.

The note becomes a navigation and argument layer rather than a warehouse of copied material. Sources remain in formats suited to reading; the note explains why they belong together.

Four practical limits of location links​

1. The link does not contain the source file​

A docube:// address stores identity and position, not the PDF or EPUB bytes. Before sending a note to another device, make sure the target document has been imported or synchronized there.

2. A custom URL needs its application​

A browser handles ordinary web links; docube:// requires DoCube. For long-term research or collaboration, write human-readable fallback metadata beside the link—title, author, page, chapter, or quotation.

3. Source changes can invalidate locations​

Replacing a PDF with another edition, rearranging EPUB chapters, or deleting Markdown headings can move the target. Preserve a short quotation and version information for important citations.

4. Sync is not backup​

If the target document is deleted, that deletion may synchronize to other devices and leave the link without a destination. Export and back up important documents and notes independently.

Use search, links, and history for different jobs​

• Search finds a source when you do not know where it is.
• Links preserve a relationship you already know matters.
• History returns through recently visited locations without advance organization.

A good knowledge workflow does not require designing every relationship up front. Search first, create links for durable connections, and let history handle temporary movement.

Frequently asked questions​

Why not copy the whole passage into the note?​

A short excerpt helps with context, but copied content separates from its source and may omit figures, footnotes, or edition details. Keep a concise quotation and an exact-location link together.

Will an EPUB link survive a font-size change?​

If the tool stores a content position rather than a displayed page number, it should return near the same passage. Replacing the EPUB or heavily changing chapter structure still requires verification.

Can linked notes replace PDF annotations?​

Not entirely. Highlights and short comments fit the immediate reading context; linked notes organize longer thinking across several sources. The two work better together.

Can a DoCube link live in another notes app?​

Yes, if the notes app preserves custom URLs and the device opening it has DoCube and the corresponding document.

If the searchable library does not exist yet, start with How to Search Across Hundreds of PDFs on Mac and How to Manage PDF and EPUB Notes Locally.

PDFs have never been a format that nobody complains about.

But I increasingly feel that many of those complaints miss the real problem.

I do not dislike PDFs. If anything, I think PDFs are still a remarkably strong document format. Their expressive range is rich, their layout is stable, and their rendering is consistent across devices. For papers, reports, court decisions, scanned books, and other layout-sensitive documents, that stability is not a flaw. It is one of the main reasons the format remains useful. Complex layout, charts, footnotes, formulas, and page relationships can all be preserved with reasonable fidelity. If the goal is to deliver content accurately to a reader, PDFs still do that job very well.

Make reuse something you can check​

Mark a useful sentence in your own PDF, write an explanation and keep its source location. After the reading session, check the record in three ways:

• Search for a word you remember and see whether you find the relevant content.
• Return from the note to the source and check whether the context supports your earlier conclusion.
• If you need to remember it, configure an appropriate review rule. Creating an annotation does not mean it becomes due immediately.

These round trips make the record useful in later reading and thinking. Bringing evidence from webpages, PDFs and videos into one note is one concrete workflow.

So I have never felt that the core problem with PDFs is that they are too old, or that they should disappear.

What bothers me more and more is not the PDF itself, but the annotation workflow built around it.

When people complain about PDFs, the first complaint is often that they are hard to edit. That is true, but for most reading, learning, and research scenarios, it is not the most important pain point. Most of the time, people are not trying to rewrite a PDF into another document. They are trying to read it, mark key passages, add comments, return a few days later, recover what mattered, compare it with other material, or quote it in their own writing.

The problem is not whether a PDF can be edited. The problem is whether what happens after annotation can keep participating in actual work.

That is where the PDF ecosystem still feels weak.

PDFs are strong at page representation, not at knowledge interaction. They are good at presenting content reliably, but not at expressing how a reader's understanding of that content should keep evolving. The PDF spec does include annotation features, but in practice they remain limited. Highlights, underlines, comments, sticky notes, and drawing tools feel more like patches attached to a page than a mature layer for working with knowledge.

As a result, almost every PDF reader ends up inventing its own annotation model.

That fact alone says a lot. The PDF ecosystem has effectively accepted that reading can be shared, but annotation cannot. The format itself takes responsibility for showing the page consistently. Once the question becomes how a reader interacts with that page, every app goes its own way. Each reader has its own highlights, its own comments, its own sync logic, and its own export format. They all work, in a limited sense, but most of them stop at a fairly primitive stage.

What does that stage usually look like? An annotation list.

You highlight twenty passages, write five comments, and in the end the reader gives you an annotation panel telling you they are all there. That is not useless. It is better than nothing. But it is also where the system often stops. It helps you look back, but rarely feeds your learning or research process in a meaningful way.

Most annotation systems in PDF readers solve the problem of recording. They do not solve the problem of continuation. Your highlights remain highlights. Your notes remain notes. They rarely enter other workflows. They do not meaningfully participate in search, in later organization, in cross-document connection, or in spaced review and reuse. They remain traces on a page rather than active objects inside a knowledge workflow.

But I increasingly feel that a valuable annotation should not be just a mark attached to a page.

It should participate in search, because annotated content is usually more important than unmarked content.

It should participate in review, because learning is rarely a one-time event.

It should participate in connection, because real understanding often emerges across multiple documents rather than inside a single one.

That is why I care less about whether a reader supports highlighting at all, and more about what happens after the highlight.

Over the last few years, while building document tools myself, I have come to believe something fairly simple: annotations should not remain side attachments to a document. They should enter the next layer of work. They should influence retrieval because marked passages are often the most valuable ones. They should enter review because important material should not end with "I highlighted this once." They should be able to connect documents because knowledge grows across references, not inside isolated files. Only then does an annotation become more than proof that I once read a passage. It becomes an entry point for future work.

That is also why I have grown increasingly impatient with another very common PDF habit: turning a PDF into a mess of handwritten notes.

I understand why people like doing this. Writing arrows, circles, stars, and margin reactions directly on a page can feel immediate and satisfying. It gives a sense of involvement. Sometimes it even creates the feeling that real thinking is happening right there on the page.

But I like this workflow less and less, for one simple reason: it barely participates in any efficient process afterward.

Handwritten notes are usually not text. They are not structured data either. They are hard to search, hard to reuse, hard to aggregate, hard to connect cleanly to other documents, and hard to bring into any more automated workflow for review, retrieval, or synthesis. In many cases, their only unquestionable value is the feeling they provide at the moment of writing. After that, they often turn into visual noise that competes with the document instead of making understanding more durable.

I am not against writing by hand, and I am not against the physical feeling of thinking through marks and gestures. What I object to is an annotation style whose value ends almost entirely with the immediate act itself. In learning and research contexts, if a note cannot enter search, review, connection, and reuse, it tends to become inefficient over time.

So what I actually want is not more decorative annotation.

I want annotations that can keep working.

They should be searchable. They should be reviewable. They should be connectable. They should be reusable. Ideally, they should preserve at least some structure instead of surviving only as visual residue.

Seen from that angle, what really needs to be rethought is not PDF reading itself. PDFs are already quite good at reading. What feels outdated is our understanding of annotation, and the role annotation should play in learning, research, and knowledge organization.

PDFs do not need to be replaced.

But the tools built around them, especially everything that happens after annotation, still need serious redesign.

Because if annotations remain only marks on a page, and cannot keep participating in search, review, connection, and reuse, then the PDF ecosystem in knowledge work still remains at a surprisingly early stage.

And the more I work on document tools, the more certain I become that this is where the next generation of them should begin.