Skip to main content
This guide takes you from an empty agents home page to a first report. You’ll create an identity for the agent to act as, create a scheduled agent from a ready-made task, and read what it hands back.

​
Before you start

  • Set up agents: enable the feature and connect a model.
  • Sign in to Console as an administrator.
  • Switch the sidebar to Agentic with the toggle at the top of the sidebar. The agents home page opens.
The agents home page, with the Agentic sidebar and the task catalogue

​
Create an identity

An agent runs using its own identity, not as you. The identity is a machine token: it has a name, an expiry and its own permissions, and an accountable Console user. You can create as many machine tokens as you like for agents, and we recommend one per agent to allow audit and control of what each agent is doing.
1

Name it and set an expiry

Go to Identities and click Create identity.Give the identity a name. It appears in the identities list and in the audit log next to everything the agent does.Acts for is you. The identity can’t hold more permissions than this user. Administrators hold every permission, so the next two steps ask you to select what the identity needs explicitly rather than inheriting everything.Choose when the token expires. The default is 90 days. When an identity expires or is revoked, its agents stop running. An agent’s identity can’t be changed afterwards, so to move to a new identity you create a new agent.The New identity step with the name, Acts for and Expires fields
2

Pre-fill the permissions from a task

On the Resources step, use Pre-fill from a task: pick the task you plan to run, such as Cluster summary, and the cluster it should run on. Console selects the minimal set of resource permissions that task needs, scoped to that one cluster. Add more resources if the agent needs them, or remove any you don’t want it to have.The Resources step with the permissions for the Cluster summary task pre-filled on one clusterThe Service access step works the same way for platform-level permissions, such as viewing the audit log or managing certificates.
3

Review and create

Check the summary and click Create.
4

Copy the token

Console shows the token once. You don’t need it to create an agent, because you pick the identity by name, but copy it if you also want to use this identity from an MCP client. It can’t be displayed again.Click Finish. The identity appears in the Identities list as Active, with the user it acts for and when it expires.The Identities list showing the new identity as Active

​
Create an agent

1

Pick the task and the model

Go to Agents > Scheduled and click Create agent.Name the agent, pick the task you created the identity for and pick the model connection that should run it. The task picker shows what the task produces, whether it runs on a schedule or on events, the permissions it needs and the tools it may use.The Task step with the Cluster summary task and a model selected
2

Pick the identity

On the Access step, select the identity you just created. The task only sees what this identity can see, and its actions are attributed to it.If you skipped the identity section, you can create one from here: pick a cluster and click Create new identity with permissions for the task. Console creates it with the permissions the task needs on that cluster.The same step sets how often the agent is allowed to run: a quiet period between runs and a maximum number of runs per hour. These guard against an event-driven agent firing on every change in a busy cluster. Leave them empty for a scheduled agent.The Access step with the identity selected and the inline identity shortcut
3

Choose when it runs

On the Condition step, pick On a schedule and a cadence: hourly, daily, weekly or monthly, with a time and a timezone. If a run is missed, for example because Console was restarted, choose Skip missed or Run once for when Console comes back.The Condition step with a daily schedule at 09:00 UTCEvent-driven agents (On an event) work the same way, except you pick which changes they react to: the resource type, the cluster, a name pattern such as orders-* and the actions that count. Console shows a preview of the audit log events from the last seven days that would have matched. Some tasks, such as Cluster summary, only run on a schedule.
4

Review and create

Check the summary. With Enable immediately on, click Create and enable: the agent runs at its next scheduled time or matching event. Turn it off to create the agent disabled and enable it later from the list or its detail page.

​
Run it and read the output

After you create the agent, Console opens its detail page. It shows the next run, the task and model, the schedule, the identity it runs as, and a History panel that fills in as it runs. You don’t have to wait for the schedule. Click Run once to start a run now. The agent detail page with the Run once button and an empty history When a run finishes, its output appears under Output in the sidebar, and the agent is listed under Activity with a link to its latest output:
  • Reports are there to read. Open one to see the result and mark it read.
  • Recommendations suggest something to do. Click Mark as actioned when you’ve done it, or Dismiss it.
  • Approvals propose a specific change, such as resetting a stuck consumer group. Approve or Reject it with an optional reason. Console records your decision and doesn’t execute the change in this release.
The Output page with the first Cluster summary report listed under Reports Every output has an Output tab with the result and a Trace tab that shows the model’s exchange with Console: which tools it called, with what arguments, and what came back. Use it to understand how the agent reached its conclusion, or why a run failed. The home page shows the next scheduled run across all agents and the most recent failure.

​
Next steps

  • Adjust the schedule once you know how long a run takes and how often the result changes.
  • Create more agents from the catalogue on the Tasks page, with one identity per agent so you can audit and control each one.
  • Disable or delete an agent from its detail page. Disabling keeps its history; deleting removes the agent but keeps the outputs it produced.