How plans and limits work

There are two kinds of limit and they behave differently.

Metered allowances

AI generations and site scans are metered. They reset at the start of each billing period.

• The tool warns you *before* it spends your last one, so nothing fails halfway through.
• A failed generation still appears in your usage ledger, because the model was called. A generation blocked by a quota check does not, because it never ran.

Standing caps

Projects, published sites and team seats are ceilings rather than allowances. They are counted live, so deleting or archiving something frees the slot immediately.

Archiving is the useful one: an archived project keeps all its scans, versions and content but stops counting against your project limit.

Hitting a limit

You get a clear message naming the limit and offering an upgrade. Nothing is lost, and nothing is deleted.

If you hit the same limit repeatedly, that is the signal to move up a tier. If it happened once in a busy week, waiting for the reset is cheaper.

Downgrading

Nothing is deleted. Projects above your new limit become read-only until you archive enough to get back under it. Published sites above the new limit are taken offline, and the versions remain.

What counts as a generation

One AI call. A site build is one generation regardless of how many pages it produces. A redesign is one. Each content item is one. Scans are counted separately, against the scan allowance.

SEO analysis and the accessibility and security audits do not consume generations at all — they are deterministic rule sets, not model calls.