PDF splitter

Taking pages out of one document

Split one PDF into smaller ones, in your browser: choose the file, state the pages you want — by typing ranges, by clicking pages in a grid, or both — pick how the selected pages should be packaged, and download the parts. The whole job is one field, one grid and one click.

• One document at a time: choosing another file replaces the first and clears the stated ranges, with a message saying so — a statement about one document is never quietly applied to another.
• The field and the grid are two hands on one selection: typing lights up the matching pages, and clicking a page rewrites the field to match.
• A grouping decides what the selection becomes: a file per range, a file per page, or everything together in one file.
• Each part downloads on its own, and one control packs them all into a single ZIP archive — two parts cost two clicks, two hundred cost one.

All of it happens on your own machine. The document is never uploaded, no server takes any part in the work, and a password typed to unlock a file goes no further than this browser tab — which, for the things people actually split, a contract, a statement, a scanned record, is the point.

Stating pages: how ranges are written

The field takes a short statement of the pages you want, written the way people already write page lists: single pages and low-to-high runs, separated by commas. 1-3,5,9-12 asks for the first three pages, then page five, then pages nine to twelve — three ranges in one line.

• A number on its own is one page; two numbers joined by a hyphen are a run, from the lower to the higher, with both ends included.
• Commas separate ranges, and whitespace around numbers and commas is ignored, so a list pasted in with spacing reads fine.
• Beyond that spacing, only digits, commas and hyphens mean anything. Any other character is turned away with its position named, so a stray letter is found rather than skipped over.
• Ranges are honoured exactly as stated: repeated pages, overlapping ranges and ranges out of order are not tidied up, because the statement is yours — under a file-per-range grouping, saying a range twice really does make two files.

Even “everything” is something you state. An empty field asks for nothing — splitting stays off — and the hint beneath it spells out the one statement that covers your document from its first page to its last.

Real mistakes are named; half-typed statements are left alone

A statement can go wrong in a handful of ways — a character that belongs to no range, a comma where a number should be — and each refusal names the position in the statement where the problem sits and says which problem it is, so you fix a character rather than re-reading the line. Three of the mistakes concern the numbers themselves, and each gets a sentence of its own.

• A range that ends before it starts, like 9-3, is refused as written. It is never silently read in the other direction, because guessing which end you meant is how a wrong page ends up in a downloaded part.
• A page past the end of the document is refused, never clamped to the last page — the tool asks you to restate it rather than deciding on its own that the overshoot did not matter.
• Page zero is refused too: pages are counted from one, so no document has a page zero to give.

A statement still being typed is not a mistake. While the field ends on a comma or a hyphen, that last gesture is simply waited out — nothing flashes red at every keystroke — but everything before it already stands on its own, so a real mistake earlier in the line is still named while you type.

The grid: every page as a clickable tile

Below the field, every page of the document is a numbered tile. Click a tile to add its page to the statement, click it again to take the page out, and the field rewrites itself to match; selection is marked with a glyph as well as a color, and the arrow keys walk the grid from the keyboard. While the field holds something the grid cannot read — a statement mid-typing, or one that was refused — the tiles go quiet instead of guessing, because a click at that moment could only throw away what you typed.

The grid describes the set of selected pages, so when it writes the field it writes the tidiest spelling of that set: ascending runs, each page named once. When your own typing said something the tidy spelling loses — 1,2,3 collapses into 1-3, and repeated, overlapping or out-of-order ranges are folded together the same way — the rewrite is announced, because under a file-per-range grouping it changes how many parts come out. Nothing but a click on a tile ever rewrites what you typed.

The three groupings: one selection, packaged three ways

Selecting pages settles what to extract; the grouping settles how many files that becomes. Take the statement 1-3,5 — two ranges, four pages — through each of the three:

• A file per range, the default, turns it into two parts: one holding the run of three pages, one holding the single page. Every range you stated becomes exactly one file.
• A file per page turns it into four parts of one page each — the grouping for feeding pages one at a time into whatever comes next.
• Everything in one file turns it into a single part holding all four pages: extraction without the separation.

The grouping changes the packaging and nothing else — the same statement selects the same pages under all three. And because ranges are honoured as stated, a range said twice really is two identical parts under the first grouping, and a page said twice really does appear twice in the everything part.

Parts, their names, and what a cut cannot bring along

Each part is an ordinary PDF, and its name is derived from your file’s so that a folder of parts stays legible: your file’s name first, then the pages the part holds. Split report.pdf at 1-3,5 with a file per range, and the downloads are report-1-3.pdf and report-5.pdf.

• Every part’s name opens with your file’s own name, less its .pdf ending, and what the scheme adds around it is plain ASCII in every language — a file name is handed on to file systems and shells, so it is never translated.
• Under a file per page each part carries its page number instead, and under everything in one file the single part carries the spelling of the whole selection.
• Two parts that would land on the same name get numbered suffixes, checked against the final list of names, so everything you download is named uniquely.

What a part holds is pages. A PDF can also carry document-level extras — a digital signature, bookmarks, form fields, an archival claim — and cutting pages out writes a new document that those extras do not follow into. When your file really carries one, a notice beside it says so before you split; the notices only ever read what the file claims about itself — that something signed it, that it declares itself archival — and verify none of it.

A locked document, and parts that are not

A document that needs a password shows a password field beside its name: type the password and unlock it right there. A wrong one is an in-place error, with the field still waiting for another try, and splitting stays off until the document opens. A file that merely restricts printing or copying is not locked in this sense — it opens by itself, with a notice that its restrictions are among the things the parts will not carry.

The password is spent on reading, never carried into the results: every part comes out unlocked, whatever the original was, and the page says so before you split. Locking the parts again is a different job for a different tool — worth knowing before you send one on believing it still protected because its source was.