> ## Documentation Index
> Fetch the complete documentation index at: https://docs.oleria.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent details

> Read an agent's details page: its owners and declared behavior, the accounts it runs as, what they can reach, who can use the agent, what it is built from, and how it appears on the Access Graph.

The agent details page puts everything Oleria knows about one AI agent in one place: who is accountable for it, which accounts it runs as, what those accounts can reach, who can use it, what it is built from, and which versions are live. Open it by selecting an agent's name in the [Agents inventory](https://docs.oleria.com/governance/agents-inventory.md).

> [!NOTE]
>
> Agent discovery is enabled per workspace. To turn it on, contact us at [support@oleria.com](mailto:support@oleria.com).

## Header

The title is the agent's display name, or its source name when it has no display name. Under it is the description from the source application, when there is one.

**Open in**, followed by the application's name, opens the agent's own page in the source application in a new tab. The button appears when the integration supplies a link for the agent.

## Details panel

The panel under the header has three columns. A dash means Oleria has no value for the field. See [How to read empty values](https://docs.oleria.com/governance/agents-inventory.md#how-to-read-empty-values).

### Governance

| Field | What it shows |
| :- | :- |
| **Owners** | The agent's owners, leaving out the ones Oleria has found aren't accountable. **None** means no owner is left. See [Owners](#owners). |
| **Lifecycle State** | The operating state the source application declares. Point to it for a one-line description. See [Lifecycle state](https://docs.oleria.com/governance/agents-inventory.md#lifecycle-state). |
| **Risk** | The agent's risk level, **No known risk**, or a dash. Point to the tag to see **Risk Details**, a numbered list of the agent's risk reasons. See [Agent risk](https://docs.oleria.com/governance/agent-risk.md). |
| **Approval State** | Where the agent stands in the source application's own approval workflow. See [Approval state](https://docs.oleria.com/governance/agents-inventory.md#approval-state). |
| **Last Activity** | The latest activity on the agent's own accounts. [Key terms](https://docs.oleria.com/governance/agents-overview.md#key-terms) defines which accounts count. |

### Identity and Source

| Field | What it shows |
| :- | :- |
| **Application** | The name of the connected instance the agent was discovered in. The inventory's **Application** column shows the application's product name instead. |
| **Type** | The source application's own name for the kind of agent record. |
| **Created By** | Who created the agent, as the source application records it. The name links to that account's page when Oleria holds it. Otherwise the source application's identifier for the creator appears. |
| **Created On** | The date the agent was created in the source application. |
| **Modified On** | The date the agent was last modified in the source application. |
| **Live Versions** | The versions of the agent that are currently live. **None** means the agent has no live version. The label reads **Live Version** when the agent has exactly one. |

### Behaviour

These fields are the source application's statements about the agent. Their values are explained under [Field values](https://docs.oleria.com/governance/agents-inventory.md#field-values).

| Field | What it shows |
| :- | :- |
| **Audience** | The widest audience the source application says may interact with the agent. |
| **Autonomy** | How independently the source application says the agent acts. The inventory labels this **Autonomy Level**. |
| **Orchestration Role** | The agent's position among other agents. |
| **Hosting Model** | Where the agent runs. |
| **Protocols** | The protocols over which the agent accepts calls. |
| **Models** | The models the agent and its live versions use, leaving out disabled ones. |

On this page, **Models** shows a dash both when the agent uses no model and when Oleria hasn't computed the list yet.

### Owners

**Owners** lists the agent's owners and leaves out every owner Oleria has found isn't an [accountable owner](https://docs.oleria.com/governance/agents-overview.md#key-terms). **None** means no owner is left, or the agent has none.

An owner Oleria hasn't evaluated yet stays listed:

* While the agent's **Risk** shows a dash, Oleria hasn't evaluated any of its owners. Every owner is listed, including an account that only created the agent.
* After that, a group stays listed while Oleria hasn't counted its members yet.

A group whose members aren't counted doesn't count as accountable. When the agent has no other accountable owner, it shows that group under **Owners** and still counts under **No Accountable Owner**, matches **Has Accountable Owner** of **False**, and has the risk reason **No one is accountable for it**.

Each name links to that owner's page when Oleria holds it. Point to a name to see why it is an owner:

| Line | Meaning |
| :- | :- |
| **Recorded as answerable by the source** | The source application names this party as answerable for the agent, without saying in what capacity. |
| **Recorded as** *capacity* **by the source** | The source application names this party in a stated capacity, for example **Recorded as Owner by the source**. The line names each capacity: **Owner**, **Manager**, **Sponsor**, **Publisher**, or **Approver**. |
| **Assigned as Owner in Oleria** | This owner was assigned in Oleria. The line appears whether or not the source application also names the owner. |
| **Created the agent** | This owner created the agent. |
| **Accountable for the agent** | Oleria evaluated this owner, and it counts as accountable. An owner Oleria hasn't evaluated yet has no such line. |

| Capacity | Meaning |
| :- | :- |
| **Owner** | Administers the agent, including its configuration and credentials. |
| **Manager** | Is designated as the agent's manager in an organizational hierarchy. |
| **Sponsor** | Vouches for the agent's business purpose and answers for its continued existence. |
| **Publisher** | Published the agent's definition. |
| **Approver** | Approved the agent's introduction or a change to it. |

## Tabs

Six tabs sit under the details panel. Each tab's label shows how many rows it holds, for example **Agent Identities (3)**. Every tab shows 10 rows per page and has its own **Manage columns** menu. On a tab with filters, the count under the table follows the filters and the count in the tab label doesn't.

Under each table, a graph shows the agent and the objects in the rows on the current page. See [Open the graph from an agent](#open-the-graph-from-an-agent).

### Agent Identities

Which accounts does this agent run as? Each row is one account the agent authenticates as.

| Column | What it shows |
| :- | :- |
| **Account** | The account's name. It links to the account's own page when Oleria holds the account. |
| **Application** | The connected instance the account belongs to. |
| **Type** | The kind of account, for example **Agent Identity**, **Service Principal**, or **Service Account** for a non-human identity (NHI), and **Standard user** or **External user** for a user account. An account with no more specific kind shows **Machine**, **Token**, or **User**. |
| **Usage** | What the agent uses the account for. **Platform** means it reaches the services of the platform the agent runs on. **Outbound** means it reaches something outside that platform, such as a third-party API or a Model Context Protocol (MCP) server. An account can show both. A dash means the source application doesn't say, or the integration doesn't read it. |
| **Privilege** | Oleria's rating of the access the account holds, based on its scopes and permissions: **Elevated**, **Standard**, or **Limited**. **None** means the account holds no access that Oleria rates. **Unknown** means it holds access whose level Oleria couldn't determine. A dash means Oleria doesn't rate this account. |
| **Shared With** | The other agents that run as the same account. **None** means no other agent does. |
| **Last Activity** | The account's own latest activity, whether or not other agents share the account. |

An agent with no rows reads **No account of its own is recorded for this agent.**

### Access

What can this agent reach? Each row is one piece of access held directly by an account the agent runs as.

| Column | What it shows |
| :- | :- |
| **Access** | What the account can reach. It links to that object, except for API access. |
| **Via Account** | The account of the agent that holds the access. |
| **Type** | The kind of thing reached: **API**, **Application**, **Resource**, or **Resource Type**. |
| **Permissions** | What the account may do with it, in Oleria's permission categories: **create**, **delete**, **execute**, **privileged**, **read**, **share**, or **update**. |
| **Scope Level** | Oleria's rating of the scopes on this access, on the same scale as the account's privilege: **Elevated**, **Standard**, **Limited**, **None**, or **Unknown**. |
| **Scopes** | The scopes on this access. |
| **Privilege** | Oleria's rating of this access, from its permissions and scopes, with the same values as **Scope Level**. |

You can filter by **Access**, **Type**, **Scope Level**, **Scopes**, and **Privilege**.

| Empty table message | Meaning |
| :- | :- |
| **No access is recorded for the accounts it runs as.** | Oleria analyzed the agent's accounts and found no access. |
| **No access matches the filters.** | Your filters exclude every row. |
| **Access appears after the agent's accounts are next analyzed.** | Oleria hasn't analyzed the agent's accounts yet. |

### Who Can Use

Who is allowed to use this agent? Each row is one grant that lets a principal use the agent through one of its entry points. Principals that the source platform itself operates aren't listed. Neither are the agent's own accounts, unless another agent also runs as them.

| Column | What it shows |
| :- | :- |
| **Principal** | The account, group, or role that holds the grant. When the principal is an account that other agents run as, a second line names them, starting with **Agent:**. |
| **Type** | The kind of principal. A group also shows its size when Oleria knows it, for example **Group · 12 members**. |
| **Via** | The entry point the grant reaches: the object people use the agent through. |
| **Access Type** | **Direct** means the grant is on the entry point itself. **Inherited from**, followed by a name, means the grant is on something that contains the entry point, such as a project, folder, or organization. **Inherited** alone means Oleria doesn't hold the name. |
| **Permission** | What the grant allows, in Oleria's permission categories. A dash means the grant carries none. |
| **Mechanism** | **Permission** means a permission the principal holds grants the access. **Assignment** means the principal is assigned to the application that is the agent's entry point. A permission can carry a second label after a dot, which names the mechanism in the source application. |

You can filter by **Principal**, **Access Type** (**Direct** or **Inherited**), **Type** (**Account**, **Group**, or **Role**), and **Mechanism** (**Assignment** or **Permission**).

An agent with no rows reads **No principal is recorded as able to invoke this agent.** When your filters exclude every row, the table reads **No principal that can invoke this agent matches the filters.**

### Composition

What is this agent built from? Each row is one object that the agent, or one of its live versions, depends on. Guardrails and observers are on the **Controls** tab instead.

| Column | What it shows |
| :- | :- |
| **Object** | The object the agent depends on. It links to the object's page when Oleria has one for it. |
| **Type** | The kind of object. |
| **Used As** | What the agent uses the object for. See the values below. |
| **Runs With** | Whose authority the agent uses when it calls the object. See the values below. |
| **Credential** | The credential the dependency runs on, when the source application names one that Oleria holds. A dash means the source application names no credential for the dependency, or names one that Oleria doesn't hold. |
| **Approval** | **Automatic** means the source application declares no approval step. **Human approval** means a person must approve each use. **Unavailable** means the source application doesn't say. |
| **State** | **Enabled** means the dependency is switched on. **Disabled** means it is configured and switched off. **Unavailable** means the source application has no such switch. |
| **Versions** | Which versions declare the dependency. **Agent** means the agent's own configuration declares it. |

| **Used As** value | Meaning |
| :- | :- |
| **Tool** | The agent invokes the object directly, such as a function or a code interpreter. |
| **Model** | The agent sends its inference requests to the object. |
| **MCP server** | The agent connects to the object as an MCP client. The server decides which tools it offers. |
| **Knowledge source** | The agent retrieves reference content from the object and doesn't write to it. |
| **Instruction source** | The agent takes instructions from the object, such as a skill or a prompt template. |
| **Memory** | The agent writes state to the object and reads it back later. |
| **Delegate** | The agent hands work to another agent. |
| **Unavailable** | The source application names the object without saying what it is used as. |

| **Runs With** value | Meaning |
| :- | :- |
| **Agent's account** | The agent uses its own account's authority. |
| **Delegation** | The agent keeps its own identity and uses part of another identity's authority, so an audit trail can name both. |
| **Impersonation** | The agent becomes another identity, so an audit trail names only that identity. |
| **Unavailable** | The source application doesn't distinguish these. |

You can filter by **Object**, **Approval**, **State**, **Runs With**, **Used As**, and **Versions**. The **Versions** filter takes one live version.

| Empty table message | Meaning |
| :- | :- |
| **No dependency is recorded for this agent.** | The agent has no dependency. |
| **No dependency is recorded for this version.** | No dependency applies to the version you selected in the **Versions** filter. |
| **No dependency matches the filters.** | Your other filters exclude every row. |

### Controls

Which guardrails and observers apply to this agent? The table has the **Object**, **Type**, **Used As**, **State**, and **Versions** columns of the **Composition** tab.

| **Used As** value | Meaning |
| :- | :- |
| **Guardrail** | A control that evaluates what the agent receives, produces, or does, and can block it. A content filter and a policy engine are examples. |
| **Observer** | A control that receives what the agent receives, produces, or does, and can't block it. A telemetry export is an example. |

You can filter by **Object**, **State**, **Used As** (**Guardrail** or **Observer**), and **Versions**. The **Versions** filter takes one live version.

| Empty table message | Meaning |
| :- | :- |
| **No guardrail or observer is recorded for this agent.** | The agent has no guardrail or observer. Oleria lists the controls the source application reports. |
| **No guardrail or observer is recorded for this version.** | No guardrail or observer applies to the version you selected in the **Versions** filter. |
| **No guardrail or observer matches the filters.** | Your other filters exclude every row. |

### Versions

Which versions are live, and can they change? Each row is one live version of the agent.

| Column | What it shows |
| :- | :- |
| **Version** | The version's name. |
| **Dependencies** | How many **Composition** rows apply to this version. Select the number to open **Composition** filtered to the version. |
| **Live On** | The aliases that point at this version. |
| **Immutable** | **Yes** means the version can't change after it is created. **No** means it can change in place. |
| **Created On** | The date the version was created in the source application. |
| **Created By** | Who created the version. |

An agent with no rows reads **No versions are recorded for this agent.**

## Access Graph nodes for agents

Where agents appear on the [Access Graph](https://docs.oleria.com/posture/explore-access-graph.md) for your workspace, the graph shows three node types for them. Each has an entry in the graph legend.

| Node | What it is |
| :- | :- |
| **Agent** | An AI agent. |
| **Agent Version** | A live version of an agent. |
| **Credential** | A credential that authenticates an account. |

### Lines from an agent

A line from an agent isn't an access grant. It records one of these facts:

* **Agent to account** - the agent runs as that account.
* **Agent to agent version** - the version is a live version of the agent.
* **Agent to something it depends on** - the object is part of the agent's composition or controls.
* **Agent to an entry point** - people use the agent through that object. Grants to use the agent are drawn to the entry point, not to the agent.
* **Agent to an owner** - the source application names that party as answerable for the agent.

Two more lines involve the agent's accounts. A line from a principal to one of the agent's accounts means the principal can act as that account. A line from a credential to an account means the credential authenticates the account.

Which lines appear depends on what you select under **Graph types**.

### Open the graph from an agent

**1. Open a tab**

On the agent details page, open any tab. The graph under the table shows the agent and the objects in the rows on the current page.

**2. Open the full graph**

Select **Explore Access Graph**, the graph icon at the top right of the graph. It opens the Access Graph with those nodes drawn and the agent selected.

To start from the Access Graph instead, search for a name and select the option that ends with **in: Agents** or **in: Credentials**. Agent versions have no search option of their own. On an agent's side panel, **See more** opens its details page.

### Side panels

Selecting a node opens its side panel.

| Node | The panel shows |
| :- | :- |
| **Agent** | **Agent info**: the agent's name and type, **Application instance**, **ID** (the agent's identifier in the source application), **Live versions** (**Live version** when there is one), and **Owners**. When more than three owners are listed, **Show all owners**, followed by their number, opens its details page. Below are the lists **Agent Identities**, **Who Can Use**, **Composition**, and **Controls**. |
| **Agent Version** | **Agent version info**: **Aliases**, **Immutable**, **Created by**, and **ID**, then **Composition** and **Controls** for that version when its agent is also on the graph. |
| **Credential** | **Credential info**: **Application instance**, **Status**, **Expiration Date**, and **ID**, then the list **Used By**. |
| Application account | A list named **Agents running as this account**. With **Access** selected under **Graph types**, also **Impersonated by**, which lists principals that can act as the account, and **Can impersonate**, which lists accounts it can act as directly. |

Each row in a list is a toggle: turn it on to draw that row's objects on the graph.

With **Access** selected under **Graph types**, the agent panel adds the list **Can impersonate this agent's accounts**. Each row names a principal and the account it can act as. This list has no tab on the agent details page.

## Contact us

For questions, contact us at [support@oleria.com](mailto:support@oleria.com).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.