Implementation Guide
Enterprise AI Inventory: Discovering the AI Systems, Agents, Models, and Tool Connections Running in Your Environment
An enterprise AI inventory is a continuously updated record of every AI application, model, embedded SaaS feature, agent, MCP server, and tool connection in use, with each asset's owner, identity, permissions, reachable data and systems, and policy status. No single discovery source finds all of it. Combine network, identity, code, and cloud sources, then use runtime observation of agent tool calls to keep the record current.
An inventory is more than a list of models
An enterprise AI inventory covers every AI application, model, embedded SaaS feature, agent, MCP server, and tool connection in use. For each asset, the record shows the owner, the identity it runs under, its permissions, the data and systems it can reach, and its policy status.
This guide defines the minimum record for each asset, compares discovery methods, and describes how to move from one-time discovery to continuous visibility.
The minimum record for each AI asset
Each asset needs at least the following in its record:
- Asset type and name, such as application, embedded feature, model, agent, MCP server, or tool integration.
- Accountable owner and the team responsible for changes and review.
- Identity the asset runs under, including service accounts, tokens, and delegated user credentials.
- Permissions granted, compared with the permissions it actually uses.
- Tools, data stores, and systems it can reach, including other agents.
- Policy status: approved, under review, or unsanctioned, plus the controls applied and the date of last review.
What each inventory record connects
| Label | Description |
|---|---|
| Asset | The model, application, agent, MCP server, or tool connection. |
| Owner | The accountable person or team. |
| Identity | The credentials or service identity it runs under. |
| Access | The tools, data, and systems it can reach. |
| Policy status | Approval state and the controls applied. |
Comparing discovery methods and their coverage gaps
No single discovery source finds all of the AI assets in an environment. Each source leaves coverage gaps, so each one is best treated as an input to the inventory rather than the inventory itself.
The sources covered in this guide are:
- Network sources
- Identity sources
- Code sources
- Cloud sources
- Runtime observation of agent tool calls, used to keep the record current
Combining sources and reconciling the results
Combine network, identity, code, and cloud sources, then reconcile the results so each asset appears once, with its owner, identity, permissions, access, and policy status recorded as described in the minimum record above.
Keeping the inventory current with runtime governance
Discovery alone produces a snapshot. Runtime observation of agent tool calls keeps the record current after the initial discovery work is done.
From one-time discovery to continuous visibility
- Discover. Use network, identity, code, and cloud sources to find AI assets.
- Reconcile. Merge the results into one record per asset, using the minimum record fields.
- Observe. Use runtime observation of agent tool calls to keep the record current.