Skip to content

Sites

TypeWhat it provisionsDocument root
wordpressWordPress, admin account, pretty permalinks, APCu object cache, the Slipstream connectorrelease root
woocommerceeverything above plus WooCommerce, its pages, and cart-fragment suppressionrelease root
phpa working placeholder index.php you replace with a deploymentrelease root
laravelthe layout with public/ as the document root, ready for a deploymentrelease/public
statica placeholder index.html; served straight from disk, no PHP poolrelease root
proxyan nginx reverse proxy to an upstream you specify; no PHP, no databasen/a

Every type except proxy gets a Unix user, directory layout and TLS. Every type except proxy and static also gets a PHP-FPM pool. wordpress, woocommerce, php and laravel get a database.

In the panel: Sites → New site. From the CLI:

Terminal window
# WordPress
slipctl sites create blog.example.com wordpress \
'{"admin_email":"you@example.com","admin_user":"you","admin_password":"strong-password","title":"My Blog"}'
# WooCommerce — arrives with WooCommerce installed and /shop, /cart, /checkout working
slipctl sites create shop.example.com woocommerce \
'{"admin_email":"you@example.com","admin_user":"you","admin_password":"strong-password","title":"My Shop"}'
# Static
slipctl sites create www.example.com static '{}'
# Reverse proxy to an app on the same box
slipctl sites create api.example.com proxy '{"proxy_upstream":"http://127.0.0.1:9000"}'
# Laravel — deploy your code afterwards
slipctl sites create app.example.com laravel '{}'
slipctl deploy <site-id> /path/to/built/app

If you omit php_version the panel uses the version it installed (8.4 on Ubuntu 24.04, 8.5 on 26.04). Asking for a version that is not installed is rejected immediately, naming what is available, rather than failing halfway through provisioning.

/srv/sites/example.com/
├── current -> releases/20260726-151844
├── releases/
│ ├── initial/
│ └── 20260726-151844/
├── shared/
│ ├── uploads/ symlinked into every release
│ ├── wp-config.php symlinked into every release
│ └── exports/ database exports land here
├── logs/
│ ├── php-error.log
│ └── db-latest.sql most recent dump, included in backups
└── tmp/
└── sessions/ this site's PHP sessions, not shared

Owned by slip-site-<id>, with the site root itself root:slip-site-<id> so the user cannot replace the directory it is jailed in.

Set these in the panel under a site’s settings, or with PUT /api/sites/{id}/config.

OptionMeaning
cache_enabledmaster switch for the full-page cache
cache_ttl_secoverride the profile’s TTL; 0 uses the profile default
object_cachepersistent object cache (APCu) — on by default for commerce

See Caching.

OptionDefault
php.memory_limit_mb256, or 512 for WooCommerce
php.upload_max_mb64
php.max_execution_seconds120

Change the PHP version with PUT /api/sites/{id}/php.

OptionMeaning
resources.memory_high_mbthe site’s memory budget; drives worker and OPcache sizing
resources.php_workersoverride the calculated pm.max_children

Leave php_workers alone unless you have measured. The default is derived from the memory budget, and on a small server raising it makes things worse: measured on a 2-core box, 6 workers gave 5.54 req/s of uncacheable rendering, 10 gave 4.29 and 16 gave 4.24, with the CPU already at 0% idle throughout. The bottleneck is CPU, not worker count, so extra workers only add context switching and memory pressure. This is the mistake behind panels that ship pm.max_children = 250 and then swap under load.

A managed database is created automatically. database.external points a site at a database elsewhere instead, in which case Slipstream does not manage it and backups do not include it.

sftp_enabled creates a chrooted SFTP account for the site’s Unix user. The chroot is the site root, there is no shell, and the account cannot traverse above its own directory. Set the password via POST /api/sites/{id}/sftp, or add SSH keys via POST /api/sites/{id}/ssh-keys.

Slipstream rewrites a site’s vhost whole every time the site is re-rendered: a settings change, a PHP settings change, a certificate issue, and the certificate renewal check every 12 hours. A rule edited into the vhost is lost at the next one. Put rules in the site’s include directory instead:

/etc/slipstream/nginx/example.com/*.conf

The HTTPS server block includes every .conf file in it. Slipstream creates the directory, owned by root with mode 0750, and never writes, changes or deletes anything inside it. To hide an export folder in uploads, for example:

Terminal window
cat > /etc/slipstream/nginx/example.com/deny-exports.conf <<'EOF'
location ~* ^/wp-content/uploads/exports/ { return 404; }
EOF
nginx -t && systemctl reload nginx

Worth knowing:

  • Test before you reload. While a broken file is there, nginx -t fails, so every site’s changes fail too. Each failed render is rolled back to the files that were on disk and the task error carries nginx’s message naming the file and line. nginx keeps serving its loaded configuration, and restarting nginx from the panel refuses until the test passes.
  • Use a regex location. The include sits after the rule that denies dotfiles and before the static-asset and PHP locations. nginx checks regex locations in the order they appear, so a location ~ or location ~* here wins over Slipstream’s for .pdf, .jpg, .php and the rest. A plain prefix location such as location /exports/ does not, because nginx prefers a matching regex location over a prefix one. Use location ^~ /exports/ or a regex. A ^~ location also skips the dotfile rule for its paths, so only serve files from one if a dotfile there is meant to be public. A second location / duplicates the vhost’s own and fails nginx -t.
  • add_header replaces, it does not add. A location of yours that sets any add_header drops the security headers Slipstream sets at server level for requests it handles. Repeat them in your location if you need them. return and deny set no headers and are unaffected.
  • HTTPS only. Port 80 only answers certificate challenges and redirects to HTTPS, so a request reaches your rules after the redirect.
  • Staging gets an empty directory. Creating a staging site does not copy these rules. Copy them by hand if staging needs them.
  • Deleting a site leaves them. They are your files. Remove the directory yourself, or a later site with the same domain picks the rules up.
  • Existing sites pick up the include at their next re-render. Sites created before this release get the include line the next time they are rendered. Saving the site’s settings does it now.
  • Keep it root-only. nginx reads these files as root when it loads its configuration, so anyone who can write here can change how every site on the server is served. Do not loosen the directory’s permissions.
ActionWhat it does
One-click loginmints a single-use, 5-minute token and logs you into wp-admin. The token is validated with a direct database read, so it works regardless of object cache.
Plugin and theme listshows installed versions and available updates
Updateupdates core, plugins or themes, then flushes this site’s object cache so the live site does not keep serving pre-update state
Object cacheenables the APCu drop-in (default) or Redis
Cache statshit rate and memory of the object cache, read from the FPM pool rather than the CLI
Warm cachecrawls the sitemap to pre-populate the page cache

POST /api/sites/{id}/migration takes a .tar.gz of the site’s files and optionally a .sql dump, both already inside the site’s own directory (upload them over SFTP). It:

  1. extracts into a new release, refusing any archive entry that escapes the target;
  2. stages wp-content/uploads separately, since uploads are shared data and not release data;
  3. imports the database, having first dumped the current one as a rollback point;
  4. promotes the release;
  5. rewrites URLs from the old domain to the new one;
  6. flushes the object cache so the live site reflects the imported database.

If any step fails, files and database are rolled back together.

Removes the vhost, the FPM pool, the cache zone, the database and user, the certificate and its renewal config, the site tree, the log directory and the Unix user. FPM is reloaded before the user is deleted, because userdel refuses while a process still holds the identity.