PDF tools

Start with the file you need at the end. Merge PDF is for one ordered document assembled from several complete files. Split PDF is for a ZIP containing every source page as an individual PDF. Extract PDF pages creates one new PDF from selected pages, while Remove PDF pages keeps everything except the selected pages. Rotate PDF changes the orientation metadata of every page in one document. Compress PDF rewrites one document with object streams and clears several document-information fields, but it is not an image-quality control.

These operations are related, yet their outputs cannot be substituted casually. Splitting a contract into single-page files does not create the same result as extracting clauses into one continuous document. Removing pages changes the surviving pagination and can break references in a table of contents. Rotating all pages can correct a consistently sideways scan but worsen a file with mixed orientations. Compression can change file structure without producing a smaller result. State the intended page set, order, orientation, container, and size requirement before choosing a tool.

Merge PDF accepts several files and appends every page from each file in the order shown by the workbench. It does not offer drag-and-drop reordering after selection, interleave pages, insert only part of a source, or create bookmarks between documents. Arrange the input selection deliberately and inspect the numbered list before processing. If only selected pages from several files belong in the final package, extract those pages separately first, verify them, then merge the approved intermediate files in the required sequence.

Split PDF takes one file and writes each page into its own PDF inside a ZIP archive. That is useful when every page needs independent handling, naming, or distribution. It does not split by a typed range, file size, bookmark, chapter, or equal page count. Extract PDF pages is the better choice when pages 2, 5, and 9 through 12 must stay together in one output. The choice between split and extract is therefore about packaging: many one-page files versus one selected-page document.

Dividing a file can remove document-level navigation and context. A standalone page may refer to an appendix, signature block, or preceding terms that are no longer present. Before distribution, confirm that the selected output is legally, operationally, and editorially meaningful on its own. Preserve the source and record the extraction rule. The tools copy pages but do not explain their meaning, identify document sections, or determine whether separating them is permitted.

Extract and Remove accept comma-separated page numbers and inclusive ranges such as 1, 3-5, 8. Reversed ranges are normalized, duplicate selections collapse, and the final selected numbers are sorted. Page numbers are one-based in the field, matching ordinary reader numbering in a simple file, while the implementation converts them to zero-based indices internally. The selection must contain at least one valid page and stay between 1 and the detected total. Remove rejects a selection that would leave no pages.

Printed page labels can differ from file positions. A report may label its cover with no number, use Roman numerals for introductory matter, then restart Arabic numbering. The workbench selects physical PDF page positions, not labels printed inside the page or custom page-label metadata. Compare the browser-reported total with a trusted viewer and map the desired content to actual positions. Test a short selection before processing a sensitive document, especially when blank covers, scanned inserts, or attachments change the count.

Removal and extraction answer opposite questions. Extracting a small approved set is easier to review when only a few pages should leave the source. Removing a few unwanted pages is faster when nearly the whole document should remain. For privacy work, prefer the formulation that makes omissions explicit and then inspect every output page. A correct numeric range does not detect content that was placed on the wrong source page, repeated accidentally, hidden in annotations, or referenced elsewhere.

Rotate PDF offers 90 degrees clockwise, 180 degrees, or 90 degrees counterclockwise. The selected angle is added to each page's existing rotation and applied to the entire document. This works well for a scan where every page has the same sideways or upside-down orientation. It is not a per-page editor. A file with portrait text, wide horizontal tables, and one inverted scan may require separate extraction, rotation in groups, and recombination, or a different editor with page-level controls.

Rotation changes page rotation metadata rather than analyzing visual content. It does not crop margins, straighten a skewed photograph, perform optical character recognition, or rotate embedded annotations independently. A page can have correct orientation metadata while the scanned pixels inside it remain tilted. Open the downloaded PDF in more than one viewer when compatibility matters, check every page type, and print a representative sheet if the final destination is paper. Do not infer correctness from the first page thumbnail alone.

Compress PDF loads the source with pdf-lib, clears title, author, subject, and keyword fields, sets the producer to PWRKIT, and saves with object streams. It does not expose an image-resolution slider, recompress scanned photographs at a chosen quality, remove every possible metadata structure, or promise a target size. A document dominated by embedded images may shrink little, stay similar, or even grow after rewriting. Compare byte size before and after rather than assuming the operation succeeded because a new file exists.

Size is only one acceptance criterion. Open the result and inspect text, images, transparency, forms, links, fonts, page dimensions, and print output that matter to the workflow. The tool is not designed as an archival conformance converter, PDF optimizer with selectable profiles, or redaction system. Clearing a few document-information fields does not prove that all personal or hidden data is gone. If disclosure risk matters, use a dedicated inspection process and verify the final binary with tools appropriate to that policy.

Each PDF tool states whether it runs in the browser or uses a private worker. Browser-labeled operations keep the source document on the device, while private-worker operations upload the selected file for temporary processing and remove the job directory afterward. Input limits and encryption checks vary by route. Encrypted, damaged, unsupported, or otherwise unreadable files are rejected. Browser memory use grows with file size and page structure, so a device with limited memory can struggle before reaching a listed limit, particularly during a large merge or split.

The workbench holds selected files and generated output in the current session. Reset clears that state, and a reload can lose an undownloaded result. Downloads are separate files and do not overwrite the selected source through the page. Local execution does not authorize handling every confidential document in a general browser. Device policy, extensions, screen capture, downloads storage, backups, and clipboard behavior remain relevant. Use an approved environment and protect the downloaded result according to its new contents.

Use a repeatable review sequence after every operation. Confirm the output type first: PDF for merge, extract, remove, rotate, and compress; ZIP for split. Count the output pages and compare them with the written plan. Inspect first, last, and boundary pages, then scan the complete thumbnail view for unexpected order, orientation, blanks, or duplicates. Search for known text where text extraction matters, open links and forms that must survive, and check file size only after content review.

For a merge, verify every source boundary and final order. For split, open several files from the ZIP, including the first and last, and confirm zero-padded filenames sort correctly. For extraction, compare each retained physical page position with the request. For removal, verify both the deleted set and continuity of what remains. For rotation, inspect mixed page types. For compression, compare representative visual details and metadata expectations. A green ready message means processing completed; it does not certify the document's business meaning.

Test in the destination application, not only the browser used for processing. A portal may enforce file-size, encryption, page-dimension, font, or conformance rules that this workbench does not know. Printing can reveal clipped content or unexpected rotation. E-signature and form workflows may depend on structures changed by page copying or rewriting. Keep the original until the receiving system accepts the result and an authorized reviewer confirms that the transformation did not remove context or invalidate a required document property.

When several operations are needed, work from copies and name each intermediate result by its completed action. A mixed-orientation packet might be separated into groups, rotated, verified, and merged. A disclosure packet might be extracted from a master file, checked for scope, compressed only if the receiving portal requires a smaller upload, then tested again. Avoid chaining steps without review because an incorrect selection early in the sequence can be harder to detect after filenames and page counts change.

Choose the step order from information loss. Page removal and extraction discard context, while compression rewrites structure and selected metadata. Merge can make source boundaries less obvious. Preserve untouched inputs and a written page map before any of those operations. If a page needs true redaction, content editing, OCR, annotation removal, signature preservation, or form repair, stop and use a tool designed for that requirement. The PDF category handles page-level assembly, separation, rotation, and a limited rewrite, not every document-governance task. Assign a reviewer who did not prepare the selection when an omitted or misplaced page could affect payment, consent, compliance, or a legal record. Independent review catches assumptions that the operator may repeat while checking their own work.