AI Inventory
ANTS builds an inventory of the AI your organization uses from several sources: the agents that send traces, the governance profile recorded for each agent, the LLM models and agent frameworks your organization has approved, and the AI tools found by Shadow AI. This page explains where each part comes from and where to manage it.
Enterprise feature. The model and framework catalog opens from Gauge AI → AI Inventory. Access to it is controlled by the inventory permission in Roles & Access.
Agents
Agents are discovered automatically. There is nothing to register by hand.
- From traces: any application that sends traces with the ANTS SDKs or OpenTelemetry appears as an agent in AI Agent Center.
- From connectors: trace connectors pull traces from:
- AWS Bedrock: Bedrock Agents through CloudWatch Logs, and Bedrock Flows.
- Google Vertex AI: through Cloud Logging.
- Microsoft 365 Copilot: Copilot prompt and response activity from your Microsoft 365 tenant, connected in organization settings.
- Microsoft Copilot Studio: agent traces, with tool calls, from your Application Insights, connected in organization settings.
Governance profile
Each agent can have a governance profile, recorded through Evidence Collection. It includes:
- Use case and scope: AI use case name, business purpose, owning team or department, end-user type, geographic scope and deployment stage.
- Risk classification: EU AI Act Risk Level (Unacceptable, High, Limited, Minimal or Not classified) and the EU AI Act Annex III category.
- Data governance: training and inference-time data sources, PII categories, whether PCI or PHI data is collected, retention and encryption.
These declared values are what Drift Alerts compare against observed behavior.
ServiceNow CMDB sync
With the ServiceNow integration connected and CMDB sync turned on, agents are synced to the ServiceNow CMDB AI function table (cmdb_ci_function_ai). When the sync interval is On discovery, each newly discovered agent is synced automatically. You can also start a sync of all agents from the ServiceNow settings, which list each agent's sync status.
LLM models
The LLM Models tab is a curated catalog of LLM models. Each model has a key (for example openai/gpt-4o) and these attributes:
| Attribute | Description |
|---|---|
| Manufacturer, Family, Version, Size | Who makes the model and which version it is |
| Status | Lifecycle status: current, deprecated or archived |
| End of Life, End of Support | Lifecycle dates, with a link to the provider's lifecycle policy where available |
| Release Date, Training Cutoff | When the model was released and its training data cutoff |
| License, Open Weights, Tier | Licensing and commercial tier |
| Context Window, Modalities | Technical specifications |
- Search by model name, key or manufacturer.
- Filter by Manufacturer, Status and Approved.
- Compare 2 to 5 models side by side.
- Open a model to see its full details and its Organization Approval setting.
Approving models
Approval is set per organization and defines which models your agents are allowed to use. Approve or unapprove a model from the table or from its details, or select up to 100 models and approve or unapprove them in one action. Each approval records who approved the model and when.
Unapproved model usage
When an agent's traces use a model that is not approved for your organization, ANTS counts it as unapproved model usage:
- In AI Agent Center, the agent's Risk tab lists the Unapproved Models it used and the number of traces, with View Traces to open them.
- From the same section you can send a summary to Slack, Jira or Linear, if those integrations are connected.
- Unapproved model usage is one of the dimensions of the agent's risk profile on the Risk Dashboard.
Agent frameworks
The Agent Frameworks tab is a catalog of agent, orchestration and evaluation frameworks. Each entry shows the framework, version, type, language, license and whether it is open source, plus its lifecycle status. Search and filter the catalog, and approve or unapprove frameworks for your organization one at a time or in bulk.
Framework approval records your organization's decision. Unlike model approval, it is not checked against traces. Planned capabilities are listed on the roadmap.
AI tools
Shadow AI keeps a registry of the AI tools people use across the organization: web AI tools, desktop apps, CLI tools, MCP servers and AI agents. Each tool has a sanction status:
| Status | Meaning |
|---|---|
| Sanctioned | Approved for use |
| Tolerated | Known, but not formally approved |
| Unsanctioned | Not reviewed, or explicitly forbidden. A tool that has been detected but not yet reviewed counts as Unsanctioned. |
See Registry and policy to review tools and set their status.
Exporting for audit
Export a compliance report as PDF or Excel from Audit & Export. Scope it by frameworks and agents, and choose whether to include evidence details, attestations and drift history.
Next steps
- AI Agent Center: every discovered agent, with its compliance and risk signals.
- Evidence Collection: record each agent's governance profile.
- Risk Dashboard: risk across all agents, including unapproved model usage.