> ## Documentation Index
> Fetch the complete documentation index at: https://docs.atlios.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Catalog agent

> Chat with the Atlios catalog agent to plan and prepare catalog cleanups.

The catalog agent is an assistant you chat with to clean up a store's catalog.
Tell it what you want in plain language, and it plans the work and prepares the
changes for you to approve.

## What it does

The agent plans the work for you, but it never changes your store on its own.
It works out the steps, prepares the changes, and puts them on a review screen.
Nothing reaches Shopify until you approve it.

It works on Shopify catalog cleanup: existing issues, safe product edits, tags,
vendors, product types, metafields, descriptions, image alt text, and selected
image-generation workflows. If you ask for something else, it lets you know it
cannot help with that.

Catalog scans are controlled backend jobs that run on schedule. The agent can
read their findings, but it cannot start a scan manually.

## What you can ask for

The agent can prepare these changes:

<AccordionGroup>
  <Accordion title="Move tags into a metafield">
    Turn a set of related tags into a proper metafield, for example
    `Gender_Male` into a `custom.gender` metafield, and drop the old tags.
  </Accordion>

  <Accordion title="Remove unwanted tags">
    Clean up junk, test, or leftover tags across the catalog.
  </Accordion>

  <Accordion title="Fix inconsistent values">
    Merge different spellings of a vendor or product type into one consistent
    value, for example `nike` and `NIKE` into `Nike`.
  </Accordion>

  <Accordion title="Rewrite descriptions">
    Rewrite product descriptions in your brand voice. Uses credits per product.
  </Accordion>

  <Accordion title="Rewrite image alt text">
    Generate fresh alt text for product images, for accessibility and SEO. Uses
    credits.
  </Accordion>

  <Accordion title="Fill in missing metafields">
    Fill the metafields your store tracks, or fix values a scan flagged as
    wrong.
  </Accordion>

  <Accordion title="Create a metafield definition">
    Set up a new metafield definition. This always asks for explicit
    confirmation first.
  </Accordion>

  <Accordion title="Triage catalog issues">
    List current problems, return export-ready data, assign an issue to a team
    member, prepare available fixes for review, or suppress an exact confirmed
    issue scope with a reason.
  </Accordion>

  <Accordion title="Prepare safe product edits">
    Change status to draft or archived, vendor, product type, tags, or supported
    product fields and metafields. Every edit is staged for review first.
  </Accordion>

  <Accordion title="Start reviewed image work">
    Inspect image templates and jobs, process template-mismatch issues, generate
    new product-image angles from existing catalog media, or regenerate one job
    item. Generation uses credits and needs explicit confirmation. The agent
    cannot upload, approve, delete, or publish image work.
  </Accordion>
</AccordionGroup>

Most actions are reversible. Generated content, like rewritten descriptions and
alt text, is not, so review it before you approve.

## Where to find it

You can reach the agent two ways:

* **[Store home](/product/dashboards)** is the agent, full screen. Opening a
  store puts you straight into a conversation about it.
* The **agent panel** follows you as you work on other pages.

Both share the same conversation, so you can pick up where you left off. The
panel hides itself on store home, so you never have two agents at once.

## Review and approval

Every catalog edit the agent prepares waits for your approval on the review
screen. Open it, check what will change and on which products, then approve or
discard. If **Apply from chat** is enabled for the store, the agent can show an
**Apply now** button for the exact prepared edit. You still confirm each action.

Issue suppression takes effect only after you confirm its exact scope and
reason. Image generation starts only after you confirm its scope and credit use;
generated output still requires human review before publishing.

<Note>
  Rewrites and alt text use AI credits. If you run out, or your plan does not
  include an action, the agent points you to upgrade. See [Billing and
  credits](/product/billing-and-credits).
</Note>