Skip to main content
Platform AI architecture

One catalog. Three AI surfaces.

Zen AI (chat in PIM), MCP (Claude, ChatGPT, Cursor), and specialist agents (Agent Control Center) all operate on the same governed tenant catalog — with your permissions, audit trails, and approval gates.

This page is for architects, integrators, and security reviewers. It describes behavior and boundaries customers need — not internal service topology.

3AI surfaces on one catalog
90+MCP tools on REST APIs
Same IAMActions run as you in your tenant

Three paths to catalog AI

SurfaceWhereBest for
Zen AIChat in the PIM UIMerchandisers — ad-hoc questions, inline drafts on products
MCPClaude, ChatGPT, Cursor, IDEsDevelopers and power users — tool-driven automation
Specialist agentsAgent Control CenterScheduled runs, missions, insights, approval queue

All three read and write through the same REST catalog APIs the PIM UI and syndication workers use — not a separate shadow database or CSV export.

Architecture overview · API-first


Zen AI (in-product chat)

Zen AI is the conversational assistant in PIM — grounded in live products, families, attributes, completeness, and channel rules.

TopicCustomer-facing detail
EntryZen AI panel on product pages and workspace
IdentityYour logged-in PIM session — same tenant and roles
Data scopeOnly catalog data your roles allow
WritesSuggestions in chat; apply manually or through workflows — not silent bulk publish
ModelsPlatform-managed inference when using CataZenta-billed AI credits (tenant policy)

Governance → Zen AI · Workflows · Security


MCP (external AI clients)

MCP connects external assistants to your tenant catalog via structured tools — the same operations documented in the MCP tools reference.

TopicDetail
Hosted URLhttps://api.catazenta.com/mcp
ProtocolStreamable HTTP for Claude.ai and ChatGPT
AuthOAuth sign-in → Bearer token on each request (same account as PIM)
Local IDEsOptional connector for Claude Desktop, Cursor, VS Code — MCP quickstart
Tools90+ named tools (search_products, quick_publish, …)

Setup → MCP authentication · Claude.ai · ChatGPT


Specialist agents (Agent Control Center)

In-platform specialist agents run autonomously — schedules, events, or missions — and emit insights plus approval proposals before catalog writes apply.

Zen AIMCPSpecialist agents
InteractionChat in PIMTool calls in external clientsAutonomous runs
PersistenceSession chatSession-driven chainsInsights + approval queue
WritesManual apply / workflowsNamed tool calls (API-auditable)Approval queue → apply
Best forOne product, one questionDeveloper automationContinuous catalog intelligence

Agent Control Center · Approvals


Identity & permissions

Every AI path respects tenant isolation and RBAC — the assistant only sees and changes what your user or integration account is allowed to.

PathIdentityTenant scope
Zen AIPIM session (JWT)Active tenant in UI
Hosted MCPOAuth Bearer tokenSame account and tenant as PIM
Local MCP connectorIntegration user or token from onboardingConfigured tenant only
Specialist agentsai_mgmt permissions in PIMTenant agent policies

Cross-tenant access is not available through any AI surface. Details → Multi-tenant design · Authentication


AI Content Studio vs Zen AI

AI Content StudioZen AI
ModeBatch / bulk jobsConversational
Best forHundreds of SKUs at onceExplore, diagnose, refine one record
OutputAttribute updates + review queueSuggestions in chat
APIAsync generation endpoints (separate from chat)In-product chat experience

Both are tenant-scoped and review-gated before publish. → Zen AI · AI enrichment


Cloud & data (summary for reviewers)

CataZenta is hosted on enterprise cloud infrastructure with tenant-isolated catalog data. For procurement and security questionnaires:

TopicSummary
Data residencyCatalog and media stored per tenant; discuss region requirements with your account team
In-product AIZen AI and specialist agents may use platform-managed models when your tenant uses platform AI credits
External MCPYour chosen model (Claude, GPT, etc.) — CataZenta exposes tools and enforces catalog permissions
TrainingCustomer catalog data is not used to train public foundation models in standard deployments
AuditAPI and approval actions are auditable; agent runs and approvals visible in ACC

Full trust materials → Security & trust · Security architecture · AI infrastructure

Need a detailed security pack for enterprise review? Contact CataZenta.


Get started

  1. Merchandisers — open Zen AI on a product with family and core attributes filled in.
  2. Integrators — complete First API call, then MCP quickstart.
  3. Catalog ops — enable agents in Agent Control Center and review Approvals.
  4. Architects — read API-first and Integrations.