One loop runs every piece of work in CatalogLift. It starts the moment your catalog changes and doesn't end until the store confirms the fix landed. It runs at catalog scale: one order drafts 50 fixes, one decision settles 200 of them, and you can hand the whole walk to an AI agent of your own. Here is each step — and the guarantees behind it.
The home page shows what each station looks like. This page shows what each one promises — and the receipt that proves it kept the promise.
The connector plugin syncs products, variants, categories, brands, attributes and media into a normalized catalog. After the first full sync, a change feed picks up edits as they happen — so findings are always about the store as it is now, not as it was at the last export.
Scanners run the full set of checks over the normalized catalog. Most are plain rules, the same every time, with no AI. 5 ask a decision model about meaning, such as whether a description is about the product its title names, and it answers with a probability, never text. Each finding carries a severity, the evidence that triggered it, and a stable identity, so the queue stays calm instead of churning.
For findings it can fix, AI drafts exactly one change — built only from facts already in your catalog, never invented. The draft is validated against a strict schema before it becomes a proposal. For opportunities, AI first judges whether a material improvement exists at all.
Either way the decision goes through the same gate, and the model never decides for itself. The proposal shows the before/after diff, the evidence and the confidence score, and right before apply the change is compared against the store's current state one more time.
An approved change is written to WooCommerce, read back immediately, and then confirmed independently: a targeted re-scan checks that the original finding is actually gone. Only then does the loop close — and the whole chain lands in change history, where eligible changes can be rolled back for 150 days after verification.
Rollback depends on the previous value and what your connector supports. Restoring an empty SKU, GTIN, regular price, brand or attribute value requires connector support for those operations. A previous non-empty brand or attribute value, a zero or negative price, or an identifier outside the supported format cannot be restored. Removing a regular price is refused while a sale price exists. A later edit can also stop a rollback; the receipt records the result.
And when the store doesn't answer? Three receipts from bad days
Most tools demo the happy path. Catalog operations live on the other one — timeouts, half-applied changes, stores edited under your feet. The loop treats failure as a state, not a surprise. Three receipts from days that did not go to plan:
Applies are idempotent. If a response is lost mid-flight, it's recovered — the change is never executed twice.
If the product changed after a proposal was drafted, that proposal can't be approved or applied until it's re-checked.
An apply that can't be confirmed becomes an explicit item to resolve — it is never silently assumed to have worked.
Create a free account and connect your store — the first scan starts on its own. What happens to each finding after that is decided by you, or by an agent you connected.
No card required. The free plan never writes to WooCommerce.