PDF Compressor

Try PDF compression and compare the result

1. Choose one local PDF of up to 20 MiB and press Compress PDF. Selection alone does not run the engine. This version checks up to 500 physical pages and a bounded static-document subset described below.
2. The tool checks the source, recompresses eligible streams and structural objects, then checks the complete candidate. Images are not downsampled or converted to JPEG, and a smaller file is not guaranteed. Cancel stops the owned work without changing your original file.
3. Read the original and candidate byte counts. Download compressed PDF is available only for a checked smaller candidate. Open and review that copy before replacing anything; an unchanged or larger candidate is reported without a compressed download.

What this lossless encoding profile changes

A PDF contains page content, images, fonts, metadata and structural objects. This profile recompresses eligible non-lossy streams and packs structural objects. It can reduce inefficient encoding without turning pages into pictures, but it cannot promise the large reductions obtained by reducing scan resolution or image quality.

Lossless encoding optimization is not byte-for-byte preservation or a universal fidelity guarantee. The rewrite changes serialization, cross-reference layout and incremental-revision history, and may raise the minimum reader version to PDF 1.5. It deliberately retains unused data and resources; it is not redaction, metadata removal or malware cleaning.

Supported files have clean PDF 1.0–1.7 or 2.0 structure, no encryption or signature indicators, and ordinary text, vector, image and font content. Static outlines, page labels, tags and Info/XMP metadata are retained. Only plain link annotations with an internal destination or a single URI/internal-go-to action are admitted; destinations are not followed by the tool.

The checked candidate must have the same physical page count and per-position interpreted page boxes, rotation and user units. These consistency checks do not prove every pixel, link, metadata item or document feature is identical. Independent implementation fixtures and your review cover different things; no PDF/A, PDF/UA, printing, accessibility or authenticity certification is claimed.

Read the size comparison, not a promised reduction

Illustrative arithmetic: 2097152 original bytes and 1572864 candidate bytes mean 2.00 MiB and 1.50 MiB, with 524288 bytes saved and a 25.00% reduction. These numbers explain the display; they are not a measured result or prediction for your file.

If a 1000000-byte source produces a checked 1200000-byte candidate, the candidate is 200000 bytes or 20.00% larger. The answer is No smaller file produced, with no compressed download. Equal byte counts are also reported honestly rather than renaming the original as a successful result.

One byte saved from 20000000 is a real improvement even though two-decimal rounding would show 0.00%. The tool instead says less than 0.01% and keeps the exact bytes visible. One MiB is exactly 1048576 bytes; percentages and MiB labels are rounded once, half up.

Supported subset and practical limits

Input is one nonempty PDF up to 20971520 bytes and 500 physical pages. A candidate must fit 41943040 bytes. Bounded structural inspection and a 60000 millisecond processing timeout can reject a file even when another viewer opens it. These limits are not a guarantee of safe total memory, speed or support on every device.

Encrypted files are excluded even when they open without a password prompt. Forms, signature or certification indicators, attachments, portfolios, optional layers, non-link annotations, executable or chained actions, extension-dependent files and files needing recovery are unsupported. There is no password, repair, flattening or silent feature-removal mode. Parsed-indicator checks are not cryptographic signature verification or proof of an unsigned history.

Already compressed images, scans and optimized PDFs may shrink little, remain the same size or grow. This version has no image-quality or target-size control. It neither promises an email/upload threshold nor removes content to make a file fit.

A current browser with working dedicated workers and the pinned engine's WebAssembly exception support is required. Failed assets, startup, allocation, processing or output checks produce no downloadable partial file. A successful download handoff does not prove that the browser saved to disk, and no agreement among all PDF readers is promised.

Local file processing and current output

This tool adds no file upload, account, saved history, document logging or file-bearing share URL. Finite engine assets load from this site, while document bytes stay inside the current browser job. New selection, cancellation, clearing and unmounting retire obsolete work and download URLs. Existing site services remain subject to site policies; local processing does not certify those services or deployment.