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

# Control who can use agents

> Give users and groups access to Conduktor Agents, identities and outputs, and see who can run, change and read what.

Console administrators can see and do everything in the agents area. Everyone else needs agent permissions, which you assign to users and groups. On top of those permissions, Console applies two rules that keep an agent from becoming a way around someone's own access:

* Every identity has an owner, and an agent runs as an identity. Only the owner of that identity, or an administrator, can make the agent run.
* The list of outputs shows titles only. To open an output and read what the agent found, you have to hold the permissions its run had.

## Assign agent permissions

Agent permissions are service permissions, so you assign them the same way as other [role-based access control (RBAC) permissions for Console services](https://docs.conduktor.io/guide/conduktor-in-production/admin/set-up-rbac.md#manage-services-permissions).

**1. Open the user or group**

Go to **Settings** > **Users** or **Groups** and open the user or group you want to change.

**2. Select the permissions**

On the **Service access** tab, find the **Agents** section and select the permissions to grant. The section only shows when agents are turned on.

**3. Save**

Click **Save**. Users who belong to several groups get the permissions of all of them.

| Permission | ID | What it allows |
| - | - | - |
| **Can view Agents** | `agentView` | Open the home, **Agents**, **Tasks** and **Models** pages, see every agent with its run history and list every output. |
| **Can manage and run Agents** | `agentManage` | Everything in **Can view Agents**, plus create, edit, enable, disable, delete and run agents, and approve or reject approvals. The [ownership rules](#understand-who-can-run-and-change-an-agent) still apply. |
| **Can view Agent identities** | `agentIdentityView` | Open the **Identities** page and see the identities you own. |
| **Can manage Agent identities** | `agentIdentityManage` | Everything in **Can view Agent identities**, plus create, revoke and delete identities you own. |

To create and run agents, a user needs both **Can manage and run Agents** and **Can manage Agent identities**.

To assign these permissions with the Console API or CLI, use the ID as a [Console permission on the `PLATFORM` resource type](https://docs.conduktor.io/guide/reference/console-reference#console-permissions).

> [!NOTE]
>
> You can't grant agent permissions to an identity. An agent can't manage agents or identities, including its own.

## Understand who can run and change an agent

You own the identities you create: the **Acts for** field of an identity names its owner. An identity can never hold more permissions than its owner, and an agent runs unattended with the permissions of its identity.

So that nobody can borrow your access through an agent, some actions need you to own the agent's identity as well as hold the permission.

| To... | You need |
| - | - |
| Create an agent | **Can manage and run Agents**, and an identity you own for the agent to run as. |
| Edit an agent | **Can manage and run Agents**, and to own the agent's identity. An agent's identity can't be changed, so create a new agent to run as a different identity. |
| Enable an agent or click **Run once** | **Can manage and run Agents**, and to own the agent's identity. Administrators can do both for any agent. |
| Disable or delete an agent | **Can manage and run Agents**. You don't have to own its identity. |
| Create an identity | **Can manage Agent identities**. You become its owner. Administrators can also create an identity that another user owns. |
| Revoke an identity | **Can manage Agent identities**, and to own the identity. Administrators can revoke any identity. Its agents stop running. |
| Delete an identity | **Can manage Agent identities**, and to own the identity. Administrators can't delete an identity they don't own. |

When you create an event-driven agent, the preview of matching audit log events also needs the **Can access the Audit log** permission.

## Understand who can read an output

Console treats listing an output and opening it as two different things:

* **The list** shows one line per output: its title, its type, the task and agent that produced it, when, and its status. It never shows what the agent found.
* **Opening an output** shows the result itself, the trace of what the agent did and, for approvals and recommendations, its actions. The result can contain data the agent read with its identity's permissions, such as topic configurations or records.

Console records the permissions an identity has when a run starts. To open the output of that run, you have to hold those permissions yourself, on the same resources, at the time you open it.

### What you see in the list

| You are | Outputs listed for you |
| - | - |
| An administrator | All of them, with their titles. |
| A user with **Can view Agents** or **Can manage and run Agents** | All of them. For an output you can't open, Console hides the title, because a title can describe what the agent found, and shows **Restricted output** in its place. |
| A user without agent permissions | Only the outputs of agents that run as an identity you own. The agents area appears for you once there is one. |

### What you can open

| You are | Outputs you can open |
| - | - |
| An administrator | All of them. |
| Anyone else | The ones whose run only used permissions that you hold now, on the same resources. This applies to the owner of the agent's identity too. |

If your permissions change after a run, an output you could read before can show **You no longer have access to this output**. Ask an administrator to restore the permissions if you need to see it.

To approve or reject an approval, you need **Can manage and run Agents** as well as being able to open it. Anyone who can open a recommendation can mark it as actioned or dismiss it.

## Related resources

* [Set up RBAC for users and groups](https://docs.conduktor.io/guide/conduktor-in-production/admin/set-up-rbac.md)
* [Set up Conduktor Agents](https://docs.conduktor.io/guide/conduktor-in-production/automate/agents/configure.md)
* [Create and run your first agent](https://docs.conduktor.io/guide/conduktor-in-production/automate/agents/get-started.md)
* [Give us feedback or request a feature](https://conduktor.io/roadmap)


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