Compression

Arweave storage is priced per byte, and HTML, JavaScript, CSS and JSON typically shrink 5-8x when compressed (a 169 MB static docs site uploads as 22 MiB). --compress compresses each file before upload and tags it with Content-Encoding; gateways return that header, and browsers decompress transparently.

ario-deploy deploy --wallet ./wallet.json --deploy-folder ./out --compress gzip

In the GitHub Action (compress needs v1.1.0 or later; pin the version, since the floating v1 tag is moved by hand and may lag):

- uses: ar-io/ar-io-deploy@v1.1.0
  with:
    deploy-key: ${{ secrets.DEPLOY_KEY }}
    deploy-folder: ./dist
    compress: gzip
    compress-exclude: 'llms*.txt,*.md'

• Prefer gzip. Gateways send the encoded bytes to every client, whether or not it asked for compression. Every browser and HTTP library understands gzip; br was ~17% smaller than gzip on a static docs site, but some non-browser clients cannot decode it.
• Formats that are already compressed are uploaded as-is: already-compressed formats (JPEG, PNG, GIF, WebP, AVIF, HEIC, WOFF/WOFF2, MP3, M4A, Ogg/Opus, MP4, WebM, and zip/gz/br/bz2/xz/zst/7z/rar archives). Other images and fonts (.svg, .ico, .ttf, .otf) are compressed. Every other file is compressed, even a tiny one gzip makes a few bytes larger, so its tags always match how it was planned.
• Exclude files meant for non-browser clients with --compress-exclude, e.g. text files that tools fetch with curl: --compress-exclude "llms*.txt,*.md". A pattern without / matches the file name in any directory.
• Gateways must label items they have not indexed yet. Right after a deploy, a gateway may serve a data item before it has indexed the item's tags. An ar-io-node without the fix for that (ar-io-node #964/#966) sends the gzip bytes with no Content-Encoding header, and browsers render garbage until the item is indexed -- or indefinitely, on a gateway that never indexes the bundle. The ar.io and Turbo gateways (turbo-gateway.com, ardrive.net, and those serving *.ar.io`) have the fix; other operators get it by upgrading. Deploy to a test undername first and load it through each gateway that matters, including through Wayfinder, which may pick any gateway.
• Deduplication still works, including --incremental. Compressed uploads are cached (and found on chain) under their own key, so turning compression on re-uploads each file once, and later deploys skip unchanged files as usual.