---
title: How do you keep your AI agent inventory in sync with the CMDB?
url: https://opencontroller.lyzr.com/blog/sync-ai-agent-inventory-with-cmdb
published: 2026-11-14
updated: 2026-11-14
author: Lyzr
---

# How do you keep your AI agent inventory in sync with the CMDB?

**Treat the agent inventory as the system of record for what an agent is and does, and the CMDB as the system of record for where it runs and who supports it. Sync one way from the inventory into the CMDB on a stable key, sync only ownership and support fields back, and reconcile on every discovery run so repeated scans update one CI instead of creating new ones.**

Most enterprises already have a CMDB, usually in ServiceNow, and an IT service management process built around it. When AI agents start appearing, the first instinct is to make each agent a configuration item. That is a good instinct. The trouble starts when two systems both think they own the record. This guide is part of our [shadow AI agent discovery guide](/blog/how-to-find-shadow-ai-agents).

## Should AI agents be configuration items at all?

Yes, for any agent that runs in production and that someone would raise an incident against. A CI gives you three things the agent inventory does not: a link to the business service it supports, a support group for incident routing, and change management. You want all three when an agent misbehaves at 2 a.m.

You do not need a CI for every prompt, test agent or developer experiment. A useful rule: create a CI when an agent has a production deployment, a named owner and at least one consumer outside its own team.

## Which CMDB classes should you use?

ServiceNow now ships an AI inventory data model with three layers. Product models describe the catalogue entry, digital assets hold ownership and governance, and CIs represent running deployments. According to ServiceNow's [AI inventory data model write-up](https://www.servicenow.com/community/cmdb-blog/understanding-the-ai-inventory-data-model-in-servicenow-product/ba-p/3546771), the main tables are:

| Layer | What it holds | Example tables |
|---|---|---|
| Product model | Catalogue definition of an AI system, model, prompt, dataset or MCP server | `cmdb_ai_system_product_model`, `cmdb_mcp_product_model` |
| Digital asset | Ownership and governance attributes | `alm_ai_system_digital_asset`, `alm_mcp_digital_asset` |
| CI | A specific running deployment | `cmdb_ci_function_ai` (SaaS or cloud-hosted AI), `cmdb_ci_appl_ai_application` (customer-managed), `cmdb_ci_function_mcp` (tool endpoints) |

Two details matter. There is no separate "AI agent" CI class; agents share the AI system tables with skills. And prompts and datasets have no CI tables in the current scope, so keep them in the agent inventory only. If you run a different CMDB, mirror the same split: one record for the agent definition, one for each deployment.

## Which fields should sync which way?

Most sync failures come from two-way sync on fields that should have one owner. Decide per field.

| Field | Owned by | Direction |
|---|---|---|
| Agent ID, name, version | Agent inventory | Inventory → CMDB |
| Platform and cloud account | Agent inventory (from discovery) | Inventory → CMDB |
| Model, tools, MCP servers | Agent inventory | Inventory → CMDB (as relationships) |
| Risk tier, last eval result | Agent inventory | Inventory → CMDB |
| Lifecycle state (active, quarantined, retired) | Agent inventory | Inventory → CMDB |
| Business service | CMDB | CMDB → inventory |
| Support group, assignment group | CMDB | CMDB → inventory |
| Change records | CMDB | Reference only |

The business owner is the one field both sides care about. Pick the identity provider as the tie-breaker, and have both systems read from it. Our guide on [assigning owners with your identity provider](/blog/assign-owners-to-ai-agents-with-your-identity-provider) covers that pattern.

## How do you stop discovery from creating duplicate CIs?

Duplicate CIs are the most common symptom of a bad sync. They appear when each discovery scan generates a new record because the identifier changed: a new version, a redeploy with a new endpoint, or the same agent found by two scanners.

Follow these steps:

1. **Use a stable correlation key.** Use the platform's native resource ID (for example the cloud resource ARN or resource ID) plus account, not the display name or endpoint URL.
2. **Model versions as attributes, not new CIs.** A new version updates the existing CI and writes a change record. Only a new deployment in a different environment creates a new CI.
3. **Run identification and reconciliation rules.** In ServiceNow, route inventory writes through the Identification and Reconciliation Engine (IRE) so the data source priority decides which value wins.
4. **Register one data source per field.** If cloud discovery and your agent inventory both write `model`, set which one has priority.
5. **Retire, don't delete.** When an agent disappears from discovery for a set period, mark the CI retired. Deleting breaks incident history.
6. **Report orphans weekly.** CIs with no matching inventory record, and inventory records with no CI, both need a human.

Our post on [fixing duplicate records in an AI agent inventory](/blog/fix-duplicate-records-in-an-ai-agent-inventory) goes deeper on matching rules.

## Who owns the record?

Split ownership the same way you split fields. The AI governance or platform team owns the agent inventory and the sync job. The CMDB or configuration manager owns class design, identification rules and data quality. The agent's business owner owns accuracy of purpose and risk tier. If an auditor asks who is accountable for a single agent, the answer should be one named person, visible in both systems.

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys across AWS, Azure and Google Cloud and keeps them in a governed registry of agents, versions, tools and MCP servers. It exposes REST APIs and webhooks, so a sync job can push registry changes into the CMDB as they happen rather than on a nightly batch. Registration does not mean runtime takeover: the registry can feed the CMDB even for agents Opencontroller only observes. See the [capability reference](/capability-reference) or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Should every AI agent become a CI?**
No. Create a CI when an agent has a production deployment, a named owner and consumers outside its own team. Keep prototypes and test agents in the agent inventory only.

**Does ServiceNow have an AI agent CI class?**
Not a dedicated one. Agents and skills share the AI system tables, and running deployments use classes such as AI Function (`cmdb_ci_function_ai`) or AI & Model Application.

**What key should I use to match agents across systems?**
Use the platform's native resource ID plus the account or tenant, never the display name or endpoint URL, which change on redeploy.

## Key facts

- Create a CMDB CI for an AI agent when it has a production deployment, a named owner and consumers outside its own team.
- ServiceNow's AI inventory data model has three layers: product models, digital assets and CIs; there is no dedicated AI agent CI class.
- ServiceNow CI classes for AI include AI Function (cmdb_ci_function_ai), AI & Model Application (cmdb_ci_appl_ai_application) and MCP Function (cmdb_ci_function_mcp).
- Sync agent definition fields one way from the agent inventory to the CMDB; sync business service and support group back.
- Use the platform resource ID plus account as the correlation key to avoid duplicate CIs.
- Opencontroller keeps a governed registry of agents, versions, tools and MCP servers and exposes REST APIs and webhooks.

## Related

- [Why does an AI agent inventory show duplicate or mismatched records, and how do you fix it?](https://opencontroller.lyzr.com/blog/fix-duplicate-records-in-an-ai-agent-inventory.md)
- [What is an AI agent registry?](https://opencontroller.lyzr.com/blog/what-is-an-ai-agent-registry.md)
- [How do you govern ServiceNow AI agents?](https://opencontroller.lyzr.com/blog/govern-servicenow-ai-agents.md)
- [More on Discovery and shadow agents](https://opencontroller.lyzr.com/blog/cluster/discovery-and-shadow-agents)
