Specialist agents
CataZenta ships ten in-platform specialist agents. Each agent has a dedicated role, default tenant policy, and optional event triggers. They are not the same as external MCP clients — see AI agents (MCP) for Claude, ChatGPT, and Cursor integration.
Agents read governed PIM data, reason over catalog context, and emit insights plus approval proposals. Catalog writes apply only after human approval (unless auto-apply is explicitly enabled for a safe agent type).
Agent catalog
| Agent | Purpose | Default enabled |
|---|---|---|
| Product enrichment | Completeness gaps, attribute quality, content improvements | Yes |
| SEO optimization | Meta titles, descriptions, SEO structure | Yes |
| Taxonomy | Category placement, classification conflicts | Yes |
| Compliance | Regulatory and policy attribute gaps | Yes |
| DPP | Digital Product Passport readiness | No |
| Syndication | Channel and marketplace readiness | No |
| Translation | Locale coverage gaps | No |
| Duplicate detection | Near-duplicate products | Yes (auto-apply eligible) |
| Variant grouping | Variant family consistency | No |
| Asset intelligence | Image and media quality, metadata | Yes |
Tenants override defaults in Agent Control Center → Agents → Configuration (enable/disable, auto-apply, confidence thresholds, blocked attributes).
How agents run
| Trigger | Example |
|---|---|
| Manual | Operator starts a run from Agents or a product workspace |
| Schedule | Cron — nightly SEO sweep across a category |
| Event | product.saved → enrichment + SEO; import_completed → duplicate detection |
| Workflow | Multi-step run — enrich a category, then syndication readiness check |
Outputs
Each run produces:
- Insights — structured findings with severity (
critical,warning,info) and type (e.g.seo_weakness,taxonomy_conflict) - Approval proposals — proposed attribute updates, category changes, or bulk actions awaiting review
Insight and approval types are constrained to governed schemas so teams can filter, bulk-resolve, and audit consistently.
Event-driven triggers (examples)
| Catalog event | Agents typically triggered |
|---|---|
| Product saved | Product enrichment, SEO optimization |
| Category changed | Taxonomy, compliance |
| Status → active/published | Compliance, DPP, syndication |
| Import completed | Duplicate detection, variant grouping |
Disabled agents are skipped per tenant policy.
Permissions
Agent Control Center requires ai_mgmt module permissions:
| Action | Permission |
|---|---|
| View runs, insights, approvals | ai_mgmt:prompts:view |
| Trigger runs, approve/reject | ai_mgmt:prompts:create |
| View/edit agent policies | ai_mgmt:configs:view / ai_mgmt:configs:update |
Next steps
- Agent Control Center — UI tour
- Prebuilt agent workflows — ten multi-agent templates in Workflows
- Approvals — review queue
- Catalog quality — findings-based health
- MCP overview — external AI clients on the same catalog