All guides

User guide

WP Optimizer — storage and backup

What the WordPress plugin inventories today, and why cleanup waits for a proven restore.

Where it runs: a plugin on your own WordPress hosting · Counts: connected sites

Read this first

It can take a backup, keep taking one on a schedule, and a site owner can restore from it. It is still not released, and it has been proven on one stack only.

Both halves of the product's name work. You start a backup from the WordPress admin — or have the site take one daily or weekly without being asked — and you restore from the same place. A drill does the whole round trip over the product's own routes: a backup taken, the site then damaged on purpose — an upload deleted, a must-use plugin deleted, a post removed, another added, the title changed — and the site restored. The deleted files came back, the post added after the backup was gone, the title was the backup's, and a visitor got a 503 with an explanation while files were being replaced.

What is missing is not a feature. It is evidence from more than one stack: Apache, MySQL 8, PHP 8.3, one WordPress version, around 6,900 files. Nothing here tells you what happens on nginx with PHP-FPM, on MariaDB, at 200,000 files, or on a site running WooCommerce and a page builder. Until that exists, this is not something to point at a business you cannot afford to lose. Email info@aigate.me if you want to follow it.

A backup reported as created is not a backup known to restore. Verified has one narrow meaning: the files still match the digests recorded for them, which catches a truncated or corrupted archive. Only a backup that has actually been restored from carries the stronger label, and the product marks that one itself — never the others, and never talks it back down on re-verification.

What exists today

A developer preview inside WordPress, under Tools, installed as AI Gate Optimizer & Backup. Every route checks for an administrator, not merely for somebody signed in.

Backups you can start. The whole WordPress root and this site's own database tables, not just the content directory. The job runs in bounded steps — 400 files or ten seconds, whichever comes first — so a request timeout cannot cut it off, and it resumes where it stopped. Its state lives beside the archive rather than in the database, so it is still readable when the database is not.

Backups that take themselves. Daily or weekly, chosen on the screen. A firing that collides with a backup or restore already running is skipped with a reason rather than queued behind it, and choosing a schedule does not set a job running under the person choosing it — the first run is an hour out. Only daily and weekly are offered, deliberately: WordPress will accept a recurrence that does not exist and then schedule nothing, which looks exactly like a working schedule.

Old backups you can delete, once a newer one has proven itself. A backup becomes a candidate for deletion only when a verified newer backup supersedes it — not because it is old, and not because there are more than some number of them, so a run of failed backups preserves history instead of erasing it. The newest verified backup cannot be deleted at any setting. Before anything happens you see both halves: what would go, and what is being kept and why. It asks you to type DELETE, refuses while a backup is running, and records every deletion with who asked for it. A file it cannot attribute to a backup it wrote itself is never touched.

A restore you can start, with the line drawn where it matters. Six phases: preflight, stage, config, promote, database, finish. Staging and verification finish before any live file is touched, so everything up to promote can be abandoned at no cost — and the screen tells you which side of that line the job is on. It names the backup, states what will happen, and asks you to type the word RESTORE. From promote onwards WP-Cron carries the job, because a half-finished restore has to be finished rather than walked away from.

A restored `wp-config.php` that fits the host it lands on. It keeps this server's database connection and the backup's table prefix, pinned to a literal when the file computed it at runtime. That is what makes a restore onto a different installation work rather than appear to.

Refusals, which are most of what a backup tool is for. Too little free space to finish. A file that changed between being hashed and being archived. A destination proven downloadable over the web, for which it prints the deny rule for Apache, nginx or Caddy. A restore whose target directories the web user cannot write — caught in preflight, before anything is touched.

A read-only storage preview: files by category — uploads, plugins, themes, core and the rest — and the database tables belonging to this site's own prefix, with their data and index sizes.

What it will not do for you

These are deliberate, and worth knowing before you trust the thing:

  • Nothing is deleted unless a person asks for it. Deleting superseded backups exists, but only

as something you start — the schedule never prunes. A site left to itself still fills its disk, and the free-space check still refuses the next backup rather than making room. It now names how much a prune would free, and refuses anyway. That refusal is the correct behaviour.

  • A deletion cannot be undone, and nothing is quarantined first. A copy held on the same disk

frees no space, which is the only reason to delete a backup at all. Everything protecting you therefore sits before the deletion: the plan you are shown, the floor beneath it, the age limit and the typed confirmation.

  • A scheduled backup still needs a visitor. WP-Cron fires when somebody requests a page, so on

a site with nobody visiting and no system scheduler the backup simply does not happen. The screen says that rather than showing you a reassuring next-run date.

  • No undo, and no selective restore. Files displaced by a restore are kept, and their location

is reported; putting them back is manual.

  • A restore needs a site that still boots, because the screen is served by WordPress. A site

that will not start is the separately distributed recovery utility's job — and that utility inspects and verifies a backup set, it does not restore.

  • The half-restored window is narrowed, not closed. Visitors get a 503 while files are being

replaced, and the login screen stays reachable. But the gate is WordPress code, and during promotion it is being replaced along with everything else.

Honest coverage

The backup resumes; the storage preview does not, so it is bounded — it stops at a file count or a time limit, whichever comes first.

When it stops early it says so, and tells you how much it looked at. A partial scan presented as a total is worse than no scan, because it is a wrong number that looks like a right one.

Two figures that are easy to confuse: the logical size a database reports is not the physical space it occupies on disk. Reclaiming the second is a separate job, and this preview does not claim it.

Why backup comes before cleanup

The order is deliberate, and it is not going to change:

  1. Prove a full backup can be restored. *Done — onto a separate WordPress with different

credentials and prefix, and in place on a damaged site, on one stack.*

  1. Only then, inventory what could be removed and show it as a preview.
  2. Only then, quarantine anything, reversibly, with the backup already in hand.

Deleting is easy to build and hard to undo. Every product that got this order wrong is a cautionary tale someone else has already written.

What it does not do yet

  • Pruning that happens on its own. Deleting superseded backups is always something a person

starts, so a site left unattended still fills up.

  • Choosing what time of day it runs, or holding a different schedule per site.
  • Deleting, quarantining or compressing anything in the site itself. What can be deleted

today is an old backup, never site content.

  • The media reference graph that a safe cleanup needs in order to tell an unused file from one that

a page builder references in a way no simple search will find.

  • Offsite copies, and an agency console across several sites.

Plans and allowances

The allowance counts connected sites — a site you attach is a thing you keep, so it does not reset each month. Nothing can connect one yet, so your usage reads nought, which is the true figure rather than a missing one. See the plans and billing guide.