Enable long-running tasks for an agent
Learn how to enable long-running tasks for an agent in Agent Builder so that the agent can process large sets of records in the background.
Long-running tasks let an agent repeat the same work across a large set of records in the background, without holding up a live chat. Agent Builder turns on long-running tasks only after an eligibility check confirms that the agent qualifies. For more information, see Automated long-running tasks.
The following roles can enable long-running tasks for an agent:
For more information, see Assign Agent Builder roles to users.
Prerequisites
Before you begin, confirm the following requirements:
You must be the author of the agent draft or an Agent Admin. Only the draft author and Agent Admins can start or cancel the eligibility check.
The agent must be in Draft state, and you must have saved the system prompt and the tools that you want the agent to use. The check evaluates the saved draft, not unsaved changes.
The system prompt and the selected tools must meet the following conditions:
The agent performs an operation on records that produces a data outcome, such as creating, updating, merging, classifying, relating, or resolving records. Text for a person to read does not qualify.
The agent repeats the same steps for each record. A confirmation step is acceptable if the system prompt also describes an unattended or automated mode.
The selected tools can read and act on records at that scale.
- Go to the Capabilities step.
- Under Capabilities, toggle on Long Running Tasks.Agent Builder saves the draft and starts the eligibility check. The panel shows the Checking eligibility status tag; the switch is unavailable, and Save & continue is unavailable until the check finishes. If the check runs longer than 3 minutes, the panel shows the Check failed status tag.Note: If you turn the switch off while the check is running, Agent Builder cancels the check.
- Review the status tag in the Long Running Tasks panel.
If the status tag is not Eligible for LRT, see Long-running task eligibility check issues.
Result
When the check passes, the Long Running Tasks panel shows the Eligible for LRT status tag, and the switch stays on. If the check does not pass, the switch returns to off, and a message in the panel explains the reason.
If you change the system prompt or the selected tools after a check passes, the switch stays on, but the panel shows a message that your instructions and tools selection changed, with a Validate again button. Select Validate again before you submit the agent for review.
Agent Builder does not submit an agent that has long-running tasks turned on when the check result is out of date, the check is still running, no check has run, or the check has failed. The agent stays in Draft state.
Verification steps
To confirm that long-running tasks are enabled, perform the following steps:
On the Capabilities step, confirm that the Long Running Tasks row in the capability list shows Enabled.
Select Long Running Tasks, and confirm that the status tag reads Eligible for LRT.