PDFWorkbenchPDF TOOLKIT

Merge · Split and extract · Rotate · Metadata strip

Four PDF jobs, none of them on a server

Drop a PDF in, pick a tool, and download the result: several files merged in the order you set, pages pulled out by range like 1-3, 5, one PDF per page packed into a ZIP, pages turned by 90, 180 or 270 degrees, or a copy with the title, author, subject, keywords, creator, producer, creation and modification dates and the XMP stream removed. Every run prints the real page count and file size on both sides of the operation, read from the files themselves.

In short

PDFWorkbench merges, splits, rotates and cleans PDF files inside your browser tab. Each page is copied into a new document on your own device, so the file is never uploaded, and the page count and file size are shown before and after every run. The metadata tool removes the eight Info fields and the XMP stream, then shows the two versions side by side.

Tools Merge · Extract · Split · Rotate · MetadataReads PDF 1.0 – 1.7, unencryptedEngine pdf-lib 1.17.1, bundled in the packageServer uploads 0 bytes, 0 requests

PDF tools

Your files

0

    No file yet. The tool reads PDFs from this device only.

    Uploaded: 0 bytes · Network requests: 0

    Merge

    Several PDFs become one, in the order of the list. Use the arrows or drag a row to change the order.

    The merged file keeps each page's own size and orientation.

    Features

    What each tool does

    Six things this page does without a server.

    MERGE

    Merge in the order you choose

    Load two or more PDFs and the list keeps their page count and size. Move a file up or down with the arrows, or drag the row; the merged document follows the list exactly. The result panel states how many files and pages went in and what came out.

    EXTRACT

    Ranges like 1-3, 5

    Pages are numbered across the whole list in order, and the tool prints that map so a range is unambiguous. Type 1-3, 5 and you get exactly those four pages in that order; overlapping or repeated numbers are folded to their first position.

    SPLIT

    One PDF per page, packed as a ZIP

    The split mode writes a single-page PDF for every page and packs them into one ZIP with no compression, so the download opens in any unarchiver. Each page is also offered as its own download button if you want just one.

    ROTATE

    90, 180 or 270 degrees

    Every page or only the pages in a range can be turned, clockwise, by 90, 180 or 270 degrees. The page's rotation value is what changes, so the content and the page size stay the same; a table lists the degrees each page had and now has.

    STRIP

    Metadata removed, not blanked

    The metadata tool builds a fresh document from the pages, so the Info dictionary and the XMP stream are not carried over at all, rather than being overwritten with empty strings. Title, author, subject, keywords, creator, producer, creation date, modification date and XMP are compared in a table.

    LOCAL

    The engine ships inside the page

    The PDF library is a file in this package, not a CDN link, and the tool makes no fetch, XHR or beacon call. You can load the page, go offline and still merge a hundred pages, then check the network panel to see that nothing left the tab.

    How it works

    Four steps, no setup

    Choose files, pick a tool, press one button, download.

    01

    Add your PDFs

    Drop files on the box or press Choose files. Each file is parsed in the tab and its page count and size appear in the list; a damaged or encrypted file is named in a message instead of being added.

    02

    Set the tool and its options

    Merge reads the list order, Extract takes a page range, Split makes one PDF per page, Rotate needs a scope and an angle, Metadata needs nothing but the file.

    03

    Press the button

    The pages are copied into a new document in memory. Nothing is compressed, transmitted or queued; a 300-page merge takes a moment on the device you are holding.

    04

    Check the table and download

    Files, pages and size are shown before and after, with the rotation or metadata table when it applies, and the output arrives as a normal download.

    Coverage

    What each tool keeps, changes and drops

    One engine, four operations. The last column is where each one stops.

    ToolWhat it changesWhat it guaranteesWhere it stops
    MergeSeveral documents become one; the order follows the listPage count is the sum of the parts, page size and orientation are kept per page, links and annotations stay with their pageBookmarks, form definitions, attachments, document-level JavaScript and the source's own metadata are not carried over
    ExtractOnly the pages you name remain, in the order you name themPage numbers are counted across the whole list and the map is printed; the output has exactly the pages you listedA page range cannot cross into pages it does not include; nothing is added
    Split, one PDF per pageEach page becomes its own file, packed in one ZIPFile count equals page count, and every file has one pageThe ZIP is stored, not compressed: it is a container, not a size saving
    RotateThe rotation value of each page, clockwiseContent and page dimensions are untouched; the degrees before and after are listed per pageA page that already carries a rotation is added to it, so 90 on a page at 270 becomes 0
    Metadata stripThe document's own descriptionTitle, author, subject, keywords, creator, producer, creation date, modification date and the XMP stream are absent from the outputText drawn on a page, file names you typed and annotations are not metadata and are not removed

    Every line here is checked against the file that comes back, with the page count, rotation and metadata read by an independent PDF reader. What is not measured is not claimed.

    Under the hood

    Pages are copied, the document is rebuilt

    No page is rasterised, so nothing is re-encoded and no text turns into a picture.

    your files→parse→copy pages→assemble→save
    Parse the cross-reference table and page tree are read in the tabCopy pages each page object, its content stream and its resources are copied into one new documentSave object streams are written fresh, so no dictionary from the source survives by accident

    Merging two PDFs is not a concatenation of bytes. Both files number their objects from 1, so the page objects have to be copied into a single new document with fresh numbers, and every reference from a copied page — its content stream, fonts, images, annotations — has to be rewritten to the new number. That is what the engine does, page by page, which is why the result opens as one normal PDF.

    Because the tool copies pages into a new document instead of editing the source, the metadata strip is a side effect of the same operation: a new document starts with no Info dictionary and no XMP stream. Fields are not blanked, they simply do not exist in the output, which is what a metadata reader reports back.

    Rotating sets the page's rotation value, the same thing a reader does when you press its rotate button. The page boxes and the content stream are untouched, so a 90 degree turn does not resample an image or move a single glyph.

    Questions

    What people ask before using it

    Is my PDF uploaded anywhere?

    No. The page loads one script and one PDF library from its own package, and the files you choose are read with the File API into the tab's memory. There is no fetch, XHR, WebSocket or beacon call in the tool, and no server address to send anything to. You can disconnect the network after the page has loaded and every tool still works.

    Which PDFs can it open?

    Unencrypted PDFs from version 1.0 to 1.7, including files with cross-reference streams and object streams, which is what most modern producers write. An encrypted file is refused with the file's name in the message rather than producing a broken result, and a file that is not a PDF at all is refused the same way.

    Does merging keep my bookmarks, links and form fields?

    Links and annotations live on a page, so they are copied with it and keep working. Bookmarks, the form definition that makes a field editable, attachments and document-level scripts belong to the document rather than to a page, and they are not carried into the new file. The page content itself is unchanged.

    What exactly does the metadata tool remove?

    The eight fields of the document information dictionary — title, author, subject, keywords, creator, producer, creation date and modification date — and the XMP metadata stream, which is where modern producers also store camera, software and editing history. Because the output is a new document, nothing else from the source's document catalog travels with it either.

    Does rotating change the page or the content?

    Only the rotation value of the page. Page width, height and the drawing instructions stay as they were, so a rotated page is the same page seen from another side. A page that already had a rotation is added to, so turning a page that was at 270 degrees by 90 leaves it at 0, which is the way a reader counts.

    Why is the output a different size from the input?

    The pages are the same, but the file around them is written again: object streams are rebuilt and the cross-reference table is written in the current form. A merge of several files usually comes out smaller than the sum of its parts because shared resources are stored once, and a metadata strip removes the XMP block and the Info dictionary as well.

    Limits

    What this page does not claim

    It is not a PDF editor

    Text cannot be changed, pages cannot be drawn on and images cannot be replaced. The four operations are merge, extract, split, rotate and metadata removal, and nothing else.

    Document-level features are dropped

    Bookmarks, form definitions, attachments, page labels, tagged-PDF structure and document scripts belong to the source document. A new document does not have them, and this page says so instead of pretending a merge is lossless.

    Encrypted files are refused

    A PDF that needs a password to open cannot be parsed in the tab. The tool names the file and stops; it does not guess, and it does not send the file anywhere to try.

    Large files are bounded by your device

    The whole document lives in the tab's memory while it is processed. A 500-page scan of photographs can need several hundred megabytes, and a phone will run out before a desktop does. Nothing is uploaded to make up the difference.

    The engine is a bundled library, not a promise

    pdf-lib 1.17.1 (MIT) is included in the package. It reads the common shapes of PDF that real producers write; a hand-made or damaged file can still fail, and the message then says which file failed rather than showing an empty result.

    Where the behaviour comes from

    • Engine — pdf-lib 1.17.1, MIT licence, shipped as assets/pdf-lib.min.js
    • Format — PDF 1.7 (ISO 32000-1:2008): page tree, cross-reference streams, object streams
    • Rotation — the /Rotate entry of a page, counted clockwise in multiples of 90
    • Metadata — the document information dictionary and the catalog's XMP /Metadata stream
    • Checks — page count, rotation and metadata read back with pypdf after every test run