Plan catalog work with AI, then review the actions that change your data.

AI assistance inside catalog workflows

Use the in-app assistant to get help in context and request supported product actions. Depending on your workspace settings, product actions can include AI enrichment, title and description edits, translations, image generation, and category assignment. Single-product controls and bulk actions make it possible to work on one record or a selected group.

AI-generated product edits are proposals: inspect the preview, warnings, language, and exact affected fields before approving a change. Generated image prompts and costs are previewed before image generation is authorized. Product and category operations use the workspace's existing catalog data and permissions.

Build a sequential plan of up to 20 tasks

The task planner stores a plan with a title, an ordered list of up to 20 supported actions, and an explicit product scope. The agent advances tasks in sequence, saves status and results, and can pause, resume, retry, or cancel a plan. One plan can target up to 200 products.

• Approval: pause for review at approval-required steps.
• Autonomous: explicitly authorize the exact saved task list and product scope for bounded execution.
• Hybrid: let eligible safe steps proceed while approval gates protect write actions.

The server enforces the action catalog and authorization. A draft does not run until explicitly started. Paid AI edits and enrichment require approval before generating a proposal and a separate approval to apply the exact generated changes in every mode. Autonomous does not mean unrestricted access or unreviewed AI writes.

Supported actions and safety boundaries

The task planner currently supports product quality checks, AI edit proposals, AI enrichment proposals, and linking selected products to an existing category. Category linking adds the selected relationship; it does not remove or replace existing category links.

The planner does not provide arbitrary tools, database queries, browser automation, publishing, deletion, or image generation as task-plan steps. Image generation is available through its separate preview-and-approval workflow when enabled. Check the current capability list in your workspace for the actions exposed by its deployed version.

Review AI usage and costs

The Billing page reports monthly AI token spend from the recorded usage ledger and separates in-app assistant, task-agent, and other AI costs. The category breakdown adds up to the same total; it does not create a second charge. Review any action-specific cost notice before authorizing paid AI work.

Images, research, and jobs that keep running

Image generation is its own preview. The assistant can prepare 1 to 3 images per product from a style instruction. Nothing is generated, and nothing is charged, until you approve that preview. Web research, when it is offered, is a source for a proposal. It does not publish the product and it does not replace the review of names, measurements, and claims.

A started plan continues after you close the assistant. Pause and resume are explicit. Cancel does not guarantee that a job already sent to the queue has stopped. A failed plan can be retried from the plan itself. Compare the stored before and after in Jobs before you apply an AI edit. Applying a proposal is a separate approval from generating it.