# Agentic AI inventory

The **Agentic AI inventory** shows the AI agents and tools that Mesh has found in
your organization: agents discovered by your connected security tools on
computers, and agents registered in a directory. It helps you answer which AI
tools are in use, whether they are approved, and who owns them.

## Open the inventory

The inventory is a card on your workspace dashboard. Its summary line reads like
"12 agents on 9 computers · 7 confirmed · 1 registered · 4 with no owner". Select
**View all** to open the full list.

If the card is empty, no connected security tool has reported AI tools running or
contacted on your computers, and no agents are registered in a directory yet.

## Search and filter the list

Use the search box to find an agent by tool, computer, or user. Select
**Filters** to narrow the list by:

- **How we know**: **Confirmed**, **Suspected**, or **Registered**.
- **Risk**.
- **Owner**: **No owner**.
- **Agent policy**: **Shadow AI**, **Suspected shadow AI**, **Approved**, or
  **Not allowed**.

Select **Show low-confidence matches** to include tools whose names also match
ordinary, non-AI traffic. The summary cards at the top count the same groups. They
overlap, so they do not add up to the total.

## Understand Shadow AI

The **Agent policy** pill tells you how each tool compares with your policy:

| Pill | Meaning |
| --- | --- |
| **Shadow AI** | Not approved in your agent policy, and it ran on this computer. |
| **Suspected shadow AI** | Not approved in your agent policy; only its website was contacted. |
| **Approved** | Approved in your agent policy. |
| **Not allowed** | Not allowed by your agent policy. |

Pills use each tool family's status and the policy's default for tools it does
not list. Custom tools and exceptions for a person or group do not change them.
A tool family whose status is limited to certain groups shows as **Shadow AI**
or **Suspected shadow AI**, whether it is approved or not allowed for those
groups.

If no agent policy is enabled yet, every AI tool detected on a computer is
labeled **Shadow AI** or **Suspected shadow AI**, because nothing is approved yet.
Registered agents carry no **Agent policy** pill. Set your policy in the
[AI Agent Gateway](/docs/admin-guide/ai-agent-gateway) settings to approve or
block tools.

## Open an agent

Select an agent to open its details, which has **Overview** and **Connectivity**
tabs.

- **Overview** shows the computer, user, and the tool that reported the agent,
  using times from the latest scan (about the last day), not all-time. It also
  lists detections and open risks, with how to fix them. When a computer is
  shared and the source did not report who used the agent, Mesh does not pick one,
  so the agent has no owner and a **Possible users on this computer** card lists
  the candidates.
- **Connectivity** is a diagram with four lanes: **Who uses it**, **The agent**,
  **What it connects to**, and **Other AI on the same computer**. The legend shows
  whether a link was **Seen on the device** or **Inferred from the tool**, and
  whether an owner is **Linked to a person** or **Owner not confirmed**. Entries
  marked **Sample data** in the legend are illustrations, not data from your
  environment.
