Propose. Approve. Publish.
Agentic PIM is how CataZenta operates modern catalogs — ten specialist agents read governed product data, propose enrichment and compliance fixes, and route every catalog write through a human approval queue before publish.
This is not AI bolted onto a legacy PIM. Agents, approvals, insights, and syndication share one catalog operating system with Zen AI, MCP, and REST APIs.
The propose → approve → publish loop
| Step | What happens | Where |
|---|---|---|
| 1. Agents propose | Enrichment, SEO, taxonomy, compliance, syndication readiness, and more | Specialist agents · ACC |
| 2. Teams review | Every proposed write lands in the approval queue | Approvals |
| 3. Approved changes apply | Governed writes update PIM; export jobs publish to channels | PIM · Syndication |
| 4. Monitor quality | Open insights and pending approvals roll into catalog health | Catalog quality |
Catalog Quality measures open agent insights and pending approvals — not PIM attribute fill rate alone. Completeness scores remain useful as publish gates; they answer a different question.
Ten specialist agents
CataZenta ships domain agents for enrichment, SEO, taxonomy, compliance, DPP, syndication, translation, duplicates, variants, and assets.
Always-on defaults
- Product enrichment
- SEO optimization
- Taxonomy
- Compliance
- Duplicate detection
- Asset intelligence
Enable per tenant
- DPP readiness
- Syndication readiness
- Translation gaps
- Variant grouping
Full agent catalog →
Triggers, outputs, event mappings, and per-agent configuration in the Agent Control Center.
Each run emits insights (findings with severity and type) and approval proposals (proposed attribute or content changes). Writes apply only after human approval — unless auto-apply is explicitly enabled for a safe agent type (e.g. duplicate detection).
Specialist agents vs MCP clients
| In-platform specialist agents | External MCP agents | |
|---|---|---|
| Where | Agent Control Center in PIM | Claude, ChatGPT, Cursor, custom clients |
| Runs | Scheduled, event-triggered, missions, manual | User-driven tool calls in chat or IDE |
| Persistence | Insights, approval queue, Catalog Quality | Session-driven tool chains you orchestrate |
| Writes | Approval queue → apply on approve | Named MCP tool calls — same governed catalog |
| Docs | Specialist agents | MCP overview · AI agents |
Both operate on the same structured catalog — families, attributes, channel rules, and audit trails.
Agent Control Center
The Agent Control Center is the PIM admin hub for Agentic PIM — enable agents, triage insights, review proposals, and monitor catalog health.
Manage
Agents · Approvals · Governance
Monitor
Activity · Observability · Cost & usage
Intelligence
Catalog Quality · Insights
Automate
Prebuilt workflows · Orchestration
Full page guide with PIM routes → Agent Control Center
Three surfaces, one catalog
| Surface | Best for |
|---|---|
| Agent Control Center | Autonomous agent runs, approvals, catalog quality monitoring |
| Zen AI | Conversational assistant on product pages in PIM |
| MCP | Claude, ChatGPT, Cursor — 90+ tools on your live tenant |
See also → AI Product Intelligence for enrichment, content, extraction, and workflow capabilities.
Start by role
Catalog / merchandising
Developers
Evaluating vs legacy PIM
Get started
- Open the Agent Control Center in PIM (requires
ai_mgmtpermissions). - Review enabled agents — start with enrichment, SEO, taxonomy, compliance.
- Run a pilot on one category; watch Activity for live execution steps.
- Triage Insights and review proposals in Approvals.
- Monitor Catalog quality weekly — findings, not fill rates alone.
- Wire MCP or REST APIs when you need automation outside the UI.
GO DEEPER