Back to Knowledge Base

How agents work

Agents are autonomous AI automations with a type, a trigger, and an action config. They process products through a queue, log every action, and run continuously via a cron dispatcher.

Agents are autonomous AI-powered automations that perform specific catalog operations without manual intervention. Each agent has a type (content, translation, pricing, validation, etc.), a trigger (on import, on schedule, on manual run), and an action configuration.

When triggered, agents process products through a queue and record their actions in the activity log. A cron-based dispatcher runs agents automatically, and a stuck-processing auto-reset ensures the pipeline never stalls.

The thirteen agent types

Configurable agents (7)

  • Content — generate descriptions, SEO titles, and feature bullets using AI Field Prompts and BrandGuard rules.
  • Translation — translate fields to target locales using glossary-aware AI, batch-processing multiple fields per call.
  • Categorization — auto-assign categories from your hierarchy with configurable confidence thresholds and auto-create mode.
  • Validation — enforce data quality via configurable rules including AI-powered semantic validation.
  • Pricing — apply markup rules and psychological rounding from cost prices.
  • Image — generate alt text and trigger Image Studio operations.
  • Publish — run publish rules, assign channel nodes, and trigger push to Shopify.

System agents (6 — always active)

  • Command Center — the conversational AI interface that routes natural-language requests to the right specialist agent.
  • Schema — detects unmapped attributes after import and suggests data model additions with near-duplicate detection.
  • Reporting — handles search queries, catalog analytics, and agent run monitoring across your entire data model.
  • Import — processes CSV and Excel uploads, auto-maps columns, and routes unknown fields to the Schema Agent.
  • Product Operations — handles direct product and variant operations: create, edit, bulk update, delete, duplicate, and assign categories through natural language.
  • Workflow — manages the agent pipeline: dispatch runs, monitor progress, skip or restart stages, and force-advance products.

Credits

Agent execution itself is free. Credits are consumed only for AI operations — a description (5 credits), a translation (2–3), a categorization (2), alt text (2). Validation and formula-based pricing consume zero credits.