Importing Azure OpenAI & Atlas Vector Search and Their Discovered Components into CMDB

[grgaikwad09]

Hi All,

We are enhancing our Service Graph Connector (SGC) app to support new AI data source types:

• Azure OpenAI
• Atlas Vector Search

This integration connects to a third-party platform. The third party is able to connect to the AI platform and perform a scan of the data source.

During the scan, AI assets are discovered from the data source and classified based on sensitivity. These are exposed as catalogs and may include:

• Models
• Datasets
• Prompts
• Vectors

The actual discovered components can vary depending on the AI platform. Our requirement is to bring these discovered AI components into ServiceNow CMDB.

Based on our analysis of CSDM 5 and AI Inventory:

• The AI Function [cmdb_ci_function_ai] class seems suitable for storing the AI data source details.
• For Azure OpenAI specifically, the Azure AI Agent [cmdb_ci_azure_ai_agent] class (which extends AI Function) appears appropriate.
• We have also reviewed AI Function, Azure AI Agent, and AI & Model Application classes as part of our analysis.

We also looked at how AI Control Tower (AICT) stores AI-related information using:

• AI Model Digital Assets
• AI Prompt Digital Assets
• AI Dataset Digital Assets
• AI System Digital Assets

However, these tables do not extend cmdb_ci. Since our SGC app uses IntegrationHub ETL to insert data and create relationships in CMDB, we cannot directly use tables that are not based on cmdb_ci.

Questions:

1. Is using the AI Function (and Azure AI Agent for Azure OpenAI) the correct approach for representing these AI data sources?
2. For catalogs (such as models, prompts, datasets, vectors, indexes, and similar items), what is the recommended way to store them so they work well with CMDB and relationships?
3. AI Discovery populates certain AI-related tables; are those tables meant only for AI Discovery, or can they also be populated via IntegrationHub ETL?

Appreciate any guidance from anyone who has worked on similar AI integrations.

Thank you!

@Jake Gillespie @marshallpar @Alex_Dundon