PDF Page Extractor

Most PDF splitters ask you to upload your document to a server. For an invoice batch, a signed contract or a medical record, that is not a small ask — and for anyone on a corporate network where uploads are blocked outright, it is not even possible.

PDF Page Extractor does the work on your machine. It makes no network calls at any point. There is no account, no registration, no telemetry and no analytics. Nothing is uploaded, so nothing needs deleting afterwards. It runs in air-gapped environments with the network cable pulled out.

Two ways to split, and they do not overlap

Split every N pages chops a whole document into equal pieces. Load a file, pick a number, extract. A 30-page file at 5 pages each gives you 6 files. At 4 pages each you get 7 files, with 2 pages in the last one. Presets cover 1 through 10, and a custom box takes anything up to 10,000.

Custom pages pulls out exactly what you list. Type 1,5-10,22-30 and you get one 16-page PDF. Tick Save each range as a separate PDF and the same expression gives you three separate files — which is how you split a book into chapters in a single pass.

You pick the method first, and only that method's fields appear. Choosing one never leaves the other quietly influencing the result.

A page range syntax that does not punish you

Spaces are ignored. The characters -, .., : and the word to all mean "range". Beyond plain numbers you can write odd, even, all, first, last, first 5, last 5, 10-last, -8 for pages 1 to 8, 12- for page 12 onward, every 3, or every 2nd page written out naturally.

Out-of-range numbers are clamped and flagged rather than rejected. Duplicates collapse. A reversed range like 20-5 is read the sensible way round. Invalid input produces a quiet hint under the box, not an error dialog you have to dismiss.

If you would rather click than type, one-click buttons cover All, None, Invert, Odd, Even, Every 2, Every 3, Every 5, First, Last, Current and + Current. The expression box and the thumbnail strip stay in sync in both directions — type 1,5-10 and the thumbnails light up; Ctrl-click a few thumbnails and the expression rewrites itself to match.

Nothing is re-rendered, so nothing degrades

There is no quality setting in this application, because there is nothing to trade away. Pages are copied, not redrawn. Original resolution is preserved. Text stays selectable and searchable. Vectors stay vector. Images keep their exact bytes. Fonts and embedded resources carry across. Page rotation is preserved, and so is the document metadata — title, author, subject, keywords, creator, producer, and both timestamps.

Bookmarks get particular care. The outline is remapped to the new page numbers, entries pointing at pages you did not select are removed, and the remaining tree is re-levelled. You do not end up with a table of contents full of dead links.

Batch that understands each file separately

Drop individual PDFs or entire folders, collected recursively. One rule applies to every file, but each document is parsed against its own page count. That means last resolves to the last page of that file, and Split every N runs to each file's own end — instead of a fixed range that would misfire the moment it hit a shorter document.

Worker threads are configurable at 1, 2, 4, 6 or 8. You can pause, resume and cancel a run, and cancelling never leaves a half-written file behind. Each file shows its own status in the queue: Queued, Running, Completed, Failed, Cancelled, Skipped or Needs password. Password-protected files are collected and re-prompted at the end rather than silently failing in the middle of the run.

A typical use: pull page 1 from 300 scanned invoices and merge the lot into one summary document.

Encrypted files, handled properly

Password-protected PDFs open with a prompt that has show/hide and a five-attempt limit. Remember for this session keeps the password in memory only — it is never written to disk, not in settings, not in logs.

Preview built for documents that are actually large

Thumbnails render lazily in the background, so a 5000-page file opens instantly rather than after a minute of grinding. The large preview zooms from 20% to 600% with a live percentage, plus fit-width and fit-page. You can rotate the view without touching the file, and page geometry is shown in millimetres alongside the page's own rotation. Spin the zoom control quickly and stale frames are discarded rather than queued up and painted late.

File names you can check before committing

Set a base name, prefix, suffix, auto-numbering and padding width. Placeholders work inside the base name: {name}, {n}, {page}, {start}, {end}, {pages}, {count}, {date} and {time}. Writing {name}_p{page} produces report_p7.pdf.

A live preview under the naming card shows the first three real file names plus a count of the rest, so you see what you are about to create. Illegal Windows characters and reserved names like CON, PRN and NUL are sanitised automatically.

Output options

Pick an output folder, which is remembered between sessions. Choose how duplicates are handled — auto-rename to … (1).pdf, overwrite, or skip. Optionally open the folder when the run finishes. You can rotate the extracted pages by 0, 90, 180 or 270 degrees, remove blank pages (detected by ink coverage, not file size), and merge the results into a single PDF. Fast-save mode skips deflate and deep garbage collection when speed matters more than file size.

Every run can be exported as a report in TXT, CSV or JSON, listing the source file, pages extracted, number of output files, time taken, output location, status and any error.

Built to not fall over

Errors are treated as values, not crashes. One unreadable file in a batch of 400 does not stop the other 399. Error messages are written for people — "the file is open in another program", not a stack trace. A rotating log file (5 × 1 MB) sits in %APPDATA%\PDFPageExtractor\logs\ and is reachable from Help → Open log folder.

Background workers never touch the interface directly. Every result travels through a queue drained by the main thread, which eliminates an entire family of intermittent crashes that plague Tkinter applications. Settings are written atomically, so a crash mid-save cannot corrupt them.

The build is covered by three automated suites: 60 engine assertions, 42 interface checks and 20 end-to-end runs.

What it doesn't do

This is version 1.0, and the list above is an exact description rather than a wish list. The following are not implemented:

• No OCR. Scanned PDFs are split page by page like any other file, but the text inside images is not read or made searchable.
• No bookmark-driven extraction. You cannot pick "Chapter 3" from the outline and have it work out the page range. Bookmarks are preserved and remapped in the output, but you select pages by number.
• No automatic chapter detection.
• No duplicate-page detection.
• No scripting CLI. The command line accepts a file path to open (PDFPageExtractor.exe "report.pdf", which is what makes Windows Open with work), but there is no way to run an extraction unattended from a script or scheduled task.
• No right-click context-menu integration in Windows Explorer.
• No hot-folder monitoring.
• No page reordering before extraction. Pages come out in document order.
• No PDF page-label support — if a document numbers its front matter i, ii, iii, you address those pages by their physical position.
• English only. The interface is translation-ready but no translations ship yet.

It also does not merge arbitrary PDFs together as a primary function, compress files, edit page content, fill forms or sign documents. It extracts and splits.

Requirements and installation

Windows 10 or 11. No installer, no admin rights and no Python on the target machine — the runtime is bundled into a single executable of roughly 45–65 MB. Download the zip, extract it anywhere, and double-click. It runs from a USB stick.

Windows SmartScreen may warn on first launch because this build is not yet code-signed. Choose More info → Run anyway. The version resource is embedded and the build is Authenticode-ready, so the warning will disappear once a certificate is in place.

Settings live in %APPDATA%\PDFPageExtractor\settings.json and remember your theme, window size, last folders, naming preferences, split method, duplicate policy, thread count and recent files. A Reset all settings action restores the defaults.

A note on licensing

The extraction engine is built on PyMuPDF, which is AGPL-licensed. The application is free to download and use. If you intend to redistribute it commercially in closed-source form, you would need a commercial licence from Artifex Software. This is an engineering note rather than legal advice.