Smol, the neural image compressor for Mac

Yes. Smol includes an MCP (Model Context Protocol) server, so AI coding tools can use Smol's local media-processing capabilities directly. Connect Claude Code or Codex from the in-app "Smol for AI" area and your agent can compress, convert, upscale, resize, and strip metadata from images, video, PDFs, and audio — all running locally on your Mac. Files are never sent to a cloud processing service.

Open Smol, go to the "Smol for AI" area in the sidebar, and follow the guided setup for your client. Smol supports Claude Code and Codex out of the box and shows a live Connected status for each once the MCP connection is active. You can also copy the full AI guide from that panel and paste it into your agent so it knows every tool and setting available.

Far more than a generic "compress" action. Through MCP an agent can run Classic compression (fast, exact, for images, PDFs, video, and audio), Neural compression (detail-first image compression guided by visual analysis), Convert (images, documents, video, audio, tables, and fonts), and Enhance (AI upscale 2× or 4×, strip metadata, automate batches). It can also choose the output format, quality target, compression mode, output location, resize and upscale settings, and batch behaviour — then read back structured results including input size, output size, and savings.

Yes — that is the point of the design. MCP connects your AI client to the Smol app running locally on your Mac, rather than uploading assets to a cloud processing service. Your images, videos, PDFs, and audio stay on-device for the entire job. This makes it usable for client work, unreleased product assets, and material under NDA that you would not put through an online tool.

General-purpose agents either shell out to whatever encoder happens to be installed or make a best guess at settings, and results drift between runs. Smol gives AI tools a dedicated local finishing layer: production-grade encoders, Neural detail retention, explicit and repeatable settings, and structured results the agent can verify. The same instruction produces the same output every time, so an agent can apply one standard across an entire project.

Drop any image into Smol — JPEG, PNG, WebP, AVIF, HEIC, or RAW. Smol compresses using professional-grade encoders that reduce file size by 50–90% while preserving visual quality. You can adjust the quality slider for full control. Everything runs locally on your Mac — no uploads, no accounts.

It depends on what you need, and we measured it. On the 24 public Kodak test images, Smol's AVIF output needed 15–21% fewer bytes than Apple's built-in AVIF encoder and 22–26% fewer than the best JPEG encoder to reach the same SSIMULACRA2 quality score, and its JPEG and WebP files are level with Squoosh's. Against ImageOptim with its lossy mode on, Smol's AVIF needed 20–22% fewer bytes than ImageOptim's JPEG at the same score. Smol also compresses PDFs, video and audio and converts files in the same window, locally, a whole folder at a time. It is not the fastest tool (a sips script is quicker) and it has no lossless mode (ImageOptim is the pick for that). Results, method and raw data: smolmac.com/benchmark.

Here is what we can show. On 5 October 2026 we compressed the 24 public Kodak test images with Smol, Apple's sips, Squoosh and Caesium, and scored every output against its original. Tuned to the same SSIMULACRA2 score, Smol's AVIF files were the smallest: 15–21% smaller than Apple's AVIF, 15% smaller than Squoosh's default AVIF and 22–26% smaller than the best JPEG output tested. Here is what it does not show. Smol was slower than sips and Caesium, Apple's JPEG encoder was 6% smaller at high quality, by the older SSIM metric Squoosh's AVIF was 4% smaller, and Smol has no lossless mode. The numbers cover Smol's standard compression, not Neural mode. The data and scripts are public at smolmac.com/benchmark, so you can check them yourself.

For making photos and web images smaller, yes, and we measured it. On 9 October 2026 we ran ImageOptim 1.9.3 and Smol on the same 24 Kodak test photos. At its default lossless settings ImageOptim made the JPEGs 8.9% smaller; Smol's default, which is lossy, made them 69.3% smaller. With ImageOptim's lossy mode on and both tuned to the same SSIMULACRA2 score, Smol's AVIF files were 20–22% smaller than ImageOptim's JPEGs and 71–76% smaller than its PNGs. ImageOptim cannot write AVIF, WebP or HEIC, convert formats or resize, and its last release was in October 2023. It is still the better choice for pixel-identical lossless optimization and for SVG, it is faster on JPEGs, and it is free. Kept as JPEG, the two produce files of about the same size. Details: smolmac.com/imageoptim-alternative.

Neural compression studies the image before encoding. Smol maps faces, text, edges, depth, and fine detail, then protects important regions while compressing less important regions harder. The result is dramatically smaller AVIF images that keep the visual detail people notice.

Smol builds an importance map before compression. Warm areas mark faces, text, key subjects, object edges, and unique detail that should stay crisp. Cool areas mark low-detail backgrounds, smooth surfaces, sky, and shadows that can be compressed more aggressively.

No. Every file is processed 100% locally on your Mac, whether it uses Apple Silicon or Intel. Your images, PDFs, videos, audio, Office files, tables, and fonts never leave your device. There are no accounts, no servers, no upload queue. Smol uses bundled native encoders that run entirely on your machine.

Drop your PDF into Smol and choose a quality preset: Tiny (72 DPI, email-ready), Small (150 DPI, sharing), Medium (300 DPI, everyday), Large (300 DPI, print quality), or Original. A 50 MB InDesign export can become 3–5 MB without visible quality loss. All compression runs locally — your PDF never leaves your Mac.

Yes. Smol compresses MP4, MOV, AVI, MKV, and WebM files using H.265 (HEVC) encoding — the same standard HandBrake uses, but with a simple drag-and-drop interface. No presets to configure, no command-line flags. Drop a video, choose High or Web quality, done. H.265 is roughly 50% more efficient than H.264 at equal visual quality.

Yes. Smol converts images, PDFs, Office files, video, audio, tables, and fonts locally on your Mac. Use it for everyday conversions like HEIC to JPEG, PNG to AVIF, DOCX to PDF, XLSX to CSV, MOV to MP4, and FLAC to MP3 without uploading files.

Yes. Smol supports local audio compression and conversion across common formats including MP3, AAC, WAV, FLAC, OGG, and ALAC. Choose the output format and quality, then process files locally without a browser-based converter.

AVIF is the most efficient image format every current browser can display. In our 24-image benchmark, AVIF files were 22–26% smaller than the best JPEG at the same SSIMULACRA2 quality score, and about 40% smaller at the same SSIM. Smol converts any image to AVIF with a single drop. If you're building for the web or optimising assets for performance, AVIF is the right choice. For compatibility with older software, JPEG or WebP are safer.

Drop your HEIC file (or folder of files) into Smol, set the output format to JPEG in settings, and compress. Smol batch-converts HEIC to JPEG, WebP, AVIF, or PNG. HEIC is the default format for iPhone photos — Smol converts them locally without uploading to any web service.

One-time purchase of $29. No subscription, no renewal, no recurring fees. Pay once and own Smol forever, including all future updates. Compare that to Smallpdf at $108/year or Acrobat Pro at $240/year — tools that charge you annually for features you use every day.

Yes. Smol has separate native downloads for Apple Silicon and 64-bit Intel Macs, with the same compression, conversion, AI upscaling, batch, Finder, clipboard, and update features. Apple Silicon requires macOS 13 or later; Intel requires macOS 14 or later. Choose the matching download after purchase.

Drop any image or PDF into Smol and enable "Strip metadata" in settings. Smol removes all EXIF, GPS, IPTC, and XMP data — camera model, GPS coordinates, author info, timestamps. You can strip metadata without recompressing, so the image content remains pixel-perfect while the identifying information is removed.

AI upscaling uses a machine learning model to enlarge images while reconstructing realistic detail — far sharper than traditional resizing. Smol runs its upscaler entirely on your Mac using the native engine for its processor. Enlarge images 2× or 4× with no internet connection, no API key, and no cloud upload required.

Yes. Drag a folder of 500+ files into Smol — images, PDFs, and videos mixed together — and Smol processes them all. Virtual scrolling means the interface stays fast even with 1,000+ files. You can watch per-file progress, see individual savings, cancel mid-batch, or re-optimise specific files after the run completes.

On 24 public test photos, Smol's default setting made high-quality JPEGs 69.3% smaller and PNGs 69.2% smaller, and converting the PNGs to AVIF made them 92.1% smaller (data at smolmac.com/benchmark). Results vary with the content. PDFs from InDesign or Acrobat often shrink 70–90% for email use. Videos encoded with H.265 are typically 40–60% smaller than the original. Smol always keeps the smaller result — if compression would increase file size, it keeps the original.

Yes — significantly. Online tools require uploading your file, waiting for server processing, and downloading the result. With large files or slow connections, that's minutes per file. Smol processes files in seconds using your Mac's CPU and GPU directly. No upload, no download, no queue. A 100 MB video that takes minutes online takes under 30 seconds in Smol.

No. Smol has no file size limits, no daily caps, and no monthly quotas. Drop a 4 GB video, a 50,000-image folder, or 200 PDFs at once — it all works. Online tools like TinyPNG cap free users at 5 MB per image, Smallpdf at 2 PDFs per day, and iLoveImg at 25 MB. Because Smol runs locally on your Mac, the only limit is your disk space. One $29 purchase covers unlimited compression forever.