What is actually available today
Be clear before anything else: this workflow is not runnable. Product Card Autofill is registered as planned rather than live, and the page shows a placeholder instead of a working form. The prompt design, the task presets, and the server-side output contract exist in the codebase, but the runtime panel is not wired up, so there is nothing to submit and nothing to receive.
That is stated here rather than buried, because a page that describes a tool you cannot use is only honest if it says so. If you came looking for something to run right now, the product listing writer is live and covers the overlapping ground of turning product facts into publishable copy.
What follows describes the intended behavior. Treat it as a design document for a feature in the queue, not a feature description of shipped software.
- Status: planned, with a placeholder in place of the tool
- The presets and prompt contract exist, the runtime does not
- Product listing writer is the live alternative today
The catalog problem it is designed for
Anyone who has loaded a product catalog into a marketplace knows the shape of the work. The supplier feed has forty attributes, the destination wants sixty, and the twenty in between have to come from a spec sheet, a PDF, or somebody memory. Meanwhile the values that did arrive are inconsistent: the same material spelled three ways, dimensions in mixed units, a color field containing a marketing name rather than a color.
That work is repetitive, mechanical, and exactly the kind of thing a language model is decent at, with one large caveat: it is also work where a plausible invented value is worse than a blank. A wrong material or a wrong dimension in a product card is a returns problem and, in some categories, a compliance problem.
The planned presets follow that shape. Filling missing marketplace fields, normalizing messy attributes into a consistent set, writing short per-attribute summaries, completing a storefront card, and producing a structured handoff for operations are the five jobs it is scoped around.
How it will work when it ships
It will be a server-side AI workflow, in the same family as the other AI tools on this site rather than the browser-only file tools. That means the product data you enter will be sent over HTTPS to a Cleanor server, forwarded as a prompt to a language model provider, and returned as structured text. It will not run on your device, and it should not be treated as a private processing step.
It will require a signed-in account with AI credits, the same as the other AI workflows, and will sit behind the same per-account rate limits and site-wide daily run and spend caps. Runs will be recorded against the account, inputs and outputs alike, which is worth knowing before feeding it a supplier price list.
The output contract is deliberately conservative. Rather than aiming for confident prose, it asks for field-ready content grouped by field name, with uncertain fields flagged briefly rather than guessed at. Whether that holds up against real catalog data is exactly the thing that has to be tested before the workflow goes live, which is why it has not.
- Server-side, not local: catalog data would be sent to a Cleanor server
- Sign-in, AI credits, rate limits, and daily caps will apply
- The prompt asks for uncertain fields to be flagged, not guessed