No—the Compress PDF tool does not upload your PDF; the source file stays in the current browser tab while it is read, compressed, assembled, and verified. The Compress PDF tool performs those document operations locally: it opens one PDF from your device, inspects embedded image objects, re-encodes only supported JPG images when a smaller candidate exists, builds a new PDF, and reopens the result before showing a download. The original file is never sent to a server as part of that workflow. This is a narrower privacy promise than saying every browser page is offline: it specifically means that the PDF-processing steps do not transmit the source document. The tool accepts a single local PDF no larger than 25 MB and has additional limits on eligible images and decoded pixels. It is designed for image-heavy files, not every PDF. A text-only document, for example, normally cannot become smaller because there may be no eligible JPG to replace. If no safe replacement is smaller, the tool returns an error and does not offer a download. That behavior is part of the design: it avoids presenting an unchanged or expanded file as a successful compression.
What stays on your device
Compress PDF follows a local file-processing workflow rather than sending the document to a remote compression service. Reading the PDF, checking embedded JPG image objects, encoding candidate images, assembling the revised PDF, and opening the output for verification all take place in the current browser tab. When a valid, smaller result is ready, the browser presents it for download. The source PDF never leaves your browser during the compression workflow.
That distinction is useful when avoiding document transfer is your main concern. The tool does not need to place your PDF on a company server or in a shared online workspace. It operates on a deliberately bounded set of eligible images instead of converting every page into a new screenshot. However, local processing does not mean that every PDF can be reduced. The final size still depends on the images inside the source file and on the PDF’s existing overhead. If the document contains text, vector artwork, forms, annotations, metadata, or image formats the tool cannot safely replace, those parts remain in the resulting structure.
For a broader look at the same local workflow and its privacy boundaries, see these browser-based PDF compression steps without uploading. The important distinction is that “processed locally” describes the document-handling process; it is not a promise that the PDF will always become smaller or that its images will remain lossless.
Which PDF images qualify for compression
The tool works at the embedded-image level rather than rasterizing whole pages. It looks for JPG image objects stored with DCTDecode, the PDF marker used for JPEG-compressed image data. To avoid changing the meaning of an image or breaking its page references, Compress PDF only processes unmasked DCTDecode images that use DeviceRGB or DeviceGray color data. Each candidate is decoded, its longest side is limited to 1,400 pixels, and it is re-encoded locally as a JPG at a quality setting of 0.70.
| PDF content | Compress PDF behavior | Result |
|---|---|---|
| Unmasked embedded JPG with DeviceRGB or DeviceGray data | Checks the image and creates a locally re-encoded candidate | Uses the candidate only when it is smaller than the original image stream |
| JPX, CMYK, Indexed, masked, ImageMask, or JPG streams with decode parameters | Skips the unsupported or potentially ambiguous case | Leaves the original image unchanged |
| Text, links, vector objects, forms, annotations, metadata, and unsupported image objects | Does not treat these as JPG replacement candidates | Remains in the PDF structure |
This narrow eligibility rule is deliberate. The tool skips formats and configurations it cannot handle confidently rather than guessing at a conversion. It also checks the byte size of every new JPG against the original stream. If the candidate is not actually smaller, the original image is retained. Consequently, a PDF can contain several eligible JPGs and still finish without a downloadable result if none of the safe replacements reduces the file size.
Compress one PDF without an upload
Use the process below when you have a scanned, photo-heavy, or otherwise JPG-heavy PDF and a visually smaller copy is acceptable. Keep the original available so you can compare the pages before replacing or sharing the original.
- Choose one local PDF up to 25 MB. Open the Compress PDF tool and select a PDF stored on your device. This version handles one file per run. Confirm that the document is at or below the 25 MB file-size limit before continuing.
- Select Compress PDF and wait while eligible embedded JPG images are checked and re-encoded locally. The tool inspects the document’s image objects, skips unsupported formats, and creates reduced candidates for supported unmasked DeviceRGB or DeviceGray JPG images. Candidate images use a maximum longest side of 1,400 pixels and JPEG quality of 0.70. Only a candidate that is smaller than its original stream is eligible for replacement.
- Review the reported input and output sizes, then download the verified PDF if it is smaller. Before exposing the download, the tool reopens the generated document with a PDF viewer and checks the result, including the page count. Review the two reported sizes before saving. If the output is not smaller, no download is offered.
After downloading, open the compressed copy and inspect representative pages, especially pages containing scans or photographs. The output can differ visually because supported images are recompressed rather than copied byte for byte. A smaller file is therefore not the same as a lossless result. If the document is important, compare it with the original and retain the original until you are satisfied with the copy.
Why a download may not appear
A missing download does not mean that the tool tried to upload the file. It usually means that the tool could not produce a smaller PDF using its supported JPG replacement method. This happens when the document is text-only, contains no safely replaceable JPG images, or contains eligible images whose re-encoded candidates are not smaller than the originals. Unsupported JPX, CMYK, Indexed, masked, ImageMask, and decode-parameter cases are also skipped.
The tool will not offer a file when no supported image becomes smaller. That is a useful safeguard because PDF overhead and unchanged content can outweigh a small saving from one image. A page-oriented optimizer may reduce file size by rewriting streams or flattening content, but that is outside this tool’s narrower contract. If your requirement is a specific target size rather than an optional image-based reduction, you would need a different workflow. If the output fails structural verification, the tool also withholds the download instead of presenting a malformed result as successful.
What the output preserves
Compress PDF retains the document’s page structure and replaces only eligible embedded JPG image payloads at their existing references. It does not rasterize every page into a single picture. The page content, page count, text, links, vector objects, forms, annotations, metadata, and unsupported image objects remain in the PDF structure. This makes the process different from tools that replace an entire document with page-sized images.
Preserving structure does not mean preserving every pixel. A supported JPG is decoded and re-encoded at the balanced quality and maximum dimension described above, so photographs, scans, or detailed graphics can look softer or show compression artifacts. Text and links remain part of the document rather than being converted into pixels, but visual quality still depends on the source images and the new encoding.
For a focused explanation of this distinction, see how Compress PDF preserves text and links. Always inspect the downloaded document before sharing it, particularly when the file contains signatures, fine print, maps, medical images, or other details where visual changes matter.
Limits to check before you start
The Compress PDF tool has several interacting limits that determine whether it can complete the task:
- One file: each run accepts one local PDF rather than a batch of documents.
- 25 MB maximum file size: the selected PDF cannot exceed this limit.
- Up to 80 eligible image objects: the number of images that can qualify for the bounded processing job is capped.
- At most 100 megapixels across decoded images: this limit applies to the combined images decoded during the operation, not just to one page.
These constraints make the tool most appropriate for scanned documents, photo-heavy reports, and similar PDFs where a smaller visual copy is useful. They also explain why a text-heavy file often cannot shrink: there may be no supported JPG payload for the tool to replace. PDF metadata and structural overhead can remain even after image reduction, so the final size depends on the source images and the document’s existing PDF structure.
Keep the original until the result has been opened and checked. The compressed version is best treated as an additional copy rather than an automatic replacement for a record copy, especially when the document must remain visually identical.
When a desktop optimizer is a better choice
Use a specialist desktop PDF optimizer when your requirements go beyond safe JPG recompression. Compress PDF does not perform arbitrary stream compression, guarantee lossless output, certify archival conformance, or manage print color. Those needs can involve different encodings, color profiles, fonts, transparency handling, and document-validation rules that this bounded browser tool deliberately avoids.
A desktop optimizer may be appropriate for professional prepress files, long-term archival packages, regulated records, or documents that must meet a fixed deliverable specification. For everyday scanned images and photo-heavy PDFs, however, the local workflow is straightforward: select one file, wait for the supported JPGs to be checked and re-encoded, compare the input and output sizes, and download the verified copy only when it is smaller.