🔒 Your files never leave this browser.
Fast, private PDF processing with no uploads. Everything runs locally in your browser.
When to use this If a PDF is too large to attach, compress it and check that the result is smaller and the text is still legible.
🔒 Your files never leave this browser.
Use this tool to resave one PDF and check whether its structure can be stored more efficiently. Keep the original. A scanned PDF may be large because it contains large images; this tool does not downsample those images or guarantee a target size.
These examples describe settings and decision rules, not measured file-processing results.
| Input or setting | Expected behavior |
|---|---|
| One PDF; any preset | Resaves PDF object streams. Presets change save scheduling, not three image-quality levels. |
| Scanned PDF with large images | Embedded image resolution is not reduced; the file may barely shrink or become larger. |
| An upload limit below the result size | No target-KB control is provided. Compare actual bytes and choose another permitted workflow. |
Record original size A and output size B in bytes. If B is less than A, reduction is (A − B) ÷ A × 100%. If B equals or exceeds A, there is no saving. Displayed reduction is clamped at zero, so 0% can also mean a larger file. Rounded MB labels may hide a difference.
Check page count, order, readable text and graphics. Search for a known phrase if searchable text matters. Keep the original if content differs or the copy cannot reopen. Signed and structurally complex documents need separate preservation checks; no losslessness, signature, accessibility or archival-conformance guarantee is made.
Rewriting PDF structure can add overhead or offer little benefit to an already efficient file. Compare actual bytes; keep the original when the copy is not useful. Repeating the same operation is not a reliable fix.
They change serialization scheduling. They do not resample images or represent independently demonstrated image-quality levels.
No. This message appears when calculated reduction is below 3%; it is a heuristic, not proof that other methods cannot help.
For loading failures, use a readable, unencrypted copy. If only some pages are required, confirm which may be omitted and extract them. If all pages are required, ask whether the limit can be raised or a separate image-optimization workflow is acceptable. Turning every page into a picture may sacrifice searchable text and vector detail.