Merge PDF files without uploading them

Very slightly, and then much smaller if you let the tool re-encode the pictures. On three seeded documents we measured — four photo pages, one photo page with body text, and one page of text only — the parts came to 2,001,057 bytes together, the lossless merge came to 2,001,198 bytes, and that is the new page tree and the catalog being written: 141 bytes for seven pages, which is nothing. Re-encoding the five pictures inside brought the same document down to 908,204 bytes, which is 54.6% smaller than the parts. So the answer to "the merged file is bigger than the originals" is yes for a moment, and no once the pictures are dealt with.

No. The files are read with the browser's own file reader and the merge happens in the page. There is no endpoint on this site that accepts a PDF, which is why there is no upload bar, no progress percentage and no queue to wait in. If you want to check the claim rather than take it, the same site publishes a measured upload test for image tools and invites re-running it.

Both, and in the order you put them in. The merge copies the objects of each document into one numbering space and joins their pages into a single page tree, so the words, the fonts and the links on each page are carried over exactly. Moving a document up or down in the list moves its pages in the result. What does not come across is the document-level bookmarks of the files you merged: those hang off the catalog of the first file, and a merged document has one catalog, not one per page range.

A merge is lossless in the sense that it does not redraw anything: your pages become pages, not images of pages. That is still not the same as a small file, because the pictures inside a merged document are still the pictures from the parts, which is where the bytes sit. Every PDF tool we could find stops at the merge and leaves you with a file that is bigger than what you started with, to be emailed through an attachment limit. Re-encoding the pictures once, on the merged file, is what makes the result smaller than the parts.

It is refused, and the refusal is the useful behaviour. An encrypted file's stream lengths cannot be read in plain text, so there is no safe way to move its objects into another document — anything written would be a file that will not open. The page says which of your files could not be read and leaves the rest to be merged, rather than quietly producing a broken document.

Usually, and the tool reports it when it happens. A file from a heavyweight editor can pack dozens of small objects — resource dictionaries, content streams — into one compressed stream, and those objects do not appear as ordinary entries. When the browser can inflate a compressed stream, this page opens it and the objects inside take part in the merge like any other. When it cannot, the stream is copied as it came and the result says so with a count, so you are never left guessing at a page that went missing.

Merging multiple PDF files at once is the everyday case here. There is no limit in the tool; the limit is your device's memory, since everything happens locally with no server to offload it to. Two documents is the usual shape of it, and five or six are comfortable on an ordinary laptop. What does bite is the size of the pictures inside them, because those are decoded on a canvas before they are re-encoded — a merge of a handful of scanned documents is a memory job, not a bandwidth one.