> ## 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.

# Agents overview

> Discover the AI agents in your connected applications and see the accounts each one runs as, what it can reach, who can use it, and who is accountable for it.

Oleria discovers the AI agents in your connected applications. Security and identity teams use these pages to answer four questions: which agents exist, what each one can do, who can reach it, and which ones to fix first.

An agent acts in your applications through the accounts it runs as, with whatever access those accounts hold. Some agents have no account of their own. They act with the access of whoever uses them, or through the credentials their tools run on. For those agents, the **Agent Identities** and **Access** tabs are empty, but the agent can still act. **Who Can Use** and **Composition** show who can use it and what it depends on, when Oleria holds that information.

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

## What you can do

* **Keep one inventory of agents** - list the agents from every connected application, with each agent's type, lifecycle state, risk, owners, and the accounts it runs as.
* **See what an agent can reach** - open an agent to see the access its accounts hold, down to the permissions and scopes.
* **See who can use an agent** - list the accounts, groups, and roles allowed to use it, and whether the right is granted directly or inherited.
* **See what an agent is built from** - the models, tools, Model Context Protocol (MCP) servers, knowledge sources, and other agents it depends on, and the guardrails and observers that apply to it.
* **Fix the riskiest agents first** - each agent has a risk level, with the reasons listed.
* **Follow an agent on the Access Graph** - agents, their versions, and credentials appear as nodes next to the accounts and resources they connect to.

## What counts as an agent

An agent is a software actor that operates with some degree of autonomy on behalf of your organization or a user. Oleria lists an agent when a connected application holds a record of it. The **Type** column shows what kind of record that is, in the application's own words.

An agent is separate from the accounts it runs as. Oleria lists those accounts in their own right, and the agent's page links to each one Oleria holds.

## Where agents come from

Oleria discovers agents through these integrations:

| Integration | **Type** values its agents show |
| :- | :- |
| [Microsoft Entra ID](https://docs.oleria.com/integrations/microsoft-entra-id.md), including Microsoft Agent 365 and Copilot Studio agents | **Agent Identity**, **Service Principal**, **Copilot Package** |
| [AWS](https://docs.oleria.com/integrations/aws-iam-and-s3.md), with Amazon Bedrock AgentCore | **Agent Runtime**, **Harness** |
| [Google Cloud Platform](https://docs.oleria.com/integrations/gcp.md) | **Reasoning Engine**, **Cloud Run Service**, **Registry Agent** |
| [Salesforce](https://docs.oleria.com/integrations/salesforce.md) Agentforce | **Agentforce Service Agent**, **Agentforce Employee Agent**, **Agentforce Orchestrator**, **Einstein Bot** |

Oleria also discovers agents in other connected applications that have them, for example [Box](https://docs.oleria.com/integrations/box.md), [ChatGPT](https://docs.oleria.com/integrations/chatgpt.md), and [Port](https://docs.oleria.com/integrations/port-io.md).

## Turn on agent discovery

**1. Ask Oleria to enable it**

Contact us at [support@oleria.com](mailto:support@oleria.com) and ask for agent discovery.

**2. Connect an application that has agents**

Connect one of the integrations above, if it isn't connected already, by following its integration page.

Reading agents can need permissions in the application that the integration didn't need before. Each integration's page lists the permissions that agent discovery uses.

**3. Open the inventory**

Go to **Agents** -> **Inventory**. In the classic navigation, open **Access Inventory** and select the **Agents** tab. Oleria reads an application's agents each time its integration syncs.

## Where to find agents

| Place | How to get there |
| :- | :- |
| Agents inventory | In the sidebar's **Governance** group, select **Agents** -> **Inventory**. In the classic navigation, open **Access Inventory** and select the **Agents** tab. |
| Agent details page | Select an agent's name in the inventory. The page is the same in both navigations. |
| Access Graph | Search for the agent's name and select the option that ends with **in: Agents**. |

## Key terms

| Term | Meaning |
| :- | :- |
| Agent | A software actor that operates with some degree of autonomy on behalf of your organization or a user, and that a connected application holds a record of. |
| Source application | The connected application an agent was discovered in. Many values on these pages are what the source application declares about the agent. |
| Agent identity | An account the agent runs as. An agent can run as several accounts, or have no account of its own. |
| Principal | An account, group, or role. |
| Who can use | The principals that are allowed to use the agent. |
| Composition | What the agent is built from: the models, tools, MCP servers, knowledge sources, and other objects it depends on. |
| Controls | The guardrails and observers that apply to the agent. A guardrail can block what the agent receives, produces, or does. An observer records it and can't block it. |
| Live version | A version of the agent that is currently live. Oleria lists live versions only. |
| Owner | An account, group, or role answerable for the agent. |
| Accountable owner | An owner who is an active person in your organization or a group with members. An owner doesn't count as accountable when it only created the agent, has left your organization, has a disabled account, is a person outside your organization, isn't a person, or is a group with no members. A creator that is also recorded as an owner counts like any other owner. |
| **Last Activity** | The latest activity on the agent's own accounts: the accounts it runs as that no other agent runs as and no other principal can act as. Principals the source platform itself operates aren't counted. The date records when an account was last used, not who used it. |
| Risk level | The highest severity among an agent's risk reasons: **Critical**, **High**, **Moderate**, or **Low**. |
| Risk reason | One specific condition that makes an agent risky, such as **Shares an account with other agents**. |

## Explore the Agents pages

> [card] [Agents inventory](https://docs.oleria.com/governance/agents-inventory.md)
>
> The summary tiles, columns, filters, and values of the list of agents.

> [card] [Agent details](https://docs.oleria.com/governance/agent-details.md)
>
> One agent's accounts, access, users, composition, controls, and versions.

> [card] [Agent risk](https://docs.oleria.com/governance/agent-risk.md)
>
> What each risk level and risk reason means, and what to check.

> [card] [Read the Access Graph](https://docs.oleria.com/posture/explore-access-graph.md)
>
> How nodes and lines on the Access Graph read, including agents.

## 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.