Inventory & findings
Use Overview to follow the relationships between people, endpoints, AI applications, services, and reported data connections. Inspect the evidence before making governance decisions.
Explore the inventory#
The estate map and inventory describe observed entities and their relationships. Select an entity to inspect its first/last-seen times, sources, attribution basis, and activity. Coverage depends on connected sources; an empty field means the source did not report it.
Models belong in findings and registry details. Application/runtime entries represent the observed application rather than creating a separate graph node for every model. Reported MCP/data-source relationships do not establish visibility into all local stdio MCP servers.
Investigate individual findings#
The Overview Findings table is an event stream: each row is an observation, with date, authorization, person, endpoint, provider/model, destination, and connector. Open a row for activity details and source evidence.
Filter by search, authorization, time range, event kind, connector, and source outcome. Filtering and direct pagination run on the server. Page sizes are 100, 250, or 500, with 100 by default. Refresh starts a fresh snapshot; paging keeps the current snapshot stable.
| Evidence | Meaning |
|---|---|
| AI usage | A supported connector recognized an AI request signature. |
| Network activity | Telemetry reached a recognized AI destination; it does not by itself prove inference. |
| Resource discovery | An agent, runtime, or resource was configured or discovered; this is not a usage count. |
Record approval in AI Registry#
Use Sanctioned, Unsanctioned, and Authorization Queue to focus a review. In AI Registry, review provider and model decisions individually or in bulk.
- A provider decision applies to that provider.
- A model decision is global across providers that offer that model.
- An undecided subject remains in the authorization queue when observed.
Built-in catalog entries have explicit defaults, including reviewed enterprise exceptions. Review those defaults against your organization's policy. Resetting decisions restores the catalog defaults.
Explore and remove evaluation data#
Administrators can add, restore, or remove examples in Settings → Evaluation workspace. The examples use fictional people and enterprise activity, real product names, and paused connectors without credentials or scheduled jobs.
Evaluation records have their own identities and ownership manifest. Restore repairs missing examples without duplicating them or resetting existing decisions. To remove them, preview the counts and enter delete evaluation data. Finish a partially failed removal before adding a new workspace.