Run a Background Agent

Create a reusable project agent, verify a manual run, and schedule recurring work with optional Slack delivery.

A Background Agent runs a saved task against your project. Start with a manual run to check the instructions and evidence, then enable a schedule when the result is useful enough to repeat.

For a question about the page you are currently viewing, use Judgment Agent. For a known per-trace condition that should trigger an action, use automations.

Prerequisites

• A Judgment project with the data the task should inspect
• Permission to create a Background Agent
• For Slack delivery, a connected Slack workspace and an available destination channel

Create and verify the agent

1. Open Background Agents > Agents and select New Agent.

2. Enter a Name, such as Daily monitoring review, and an optional Description that tells teammates what the agent owns.

3. Add Instructions that define the evidence to inspect, the time window, and the expected result. For example:

   Review this project's traces from the last 24 hours. Identify the three
   most important recurring failures, cite representative traces, and explain
   why each matters. If the evidence is insufficient, say so. Do not change data.

4. Leave Scheduled trigger off for the first run. Choose an Agent mode: Fast for a lightweight pass or Deep Research for more searching and reasoning.

5. Choose an Approval mode. Require approval asks before write tools create, update, or delete data. Auto approve permits those tools to run without approval prompts; organization roles and tool validation still apply.

6. Select Create, then Run now on the agent’s page.

7. Inspect the run’s conversation and final answer. Follow its citations and confirm that the findings match the requested project and time window.

These approval labels belong to Background Agent settings. The chat controls use Ask for writes and Auto-allow writes for the corresponding choices.

• To change future runs, edit the instructions in Settings and select Save.
• To direct one run, enter a prompt in the agent’s composer. This leaves the saved instructions unchanged.

Enable a schedule

• Save the agent’s instructions, then enable Settings > Triggers > Scheduled trigger.
• Choose a schedule below and Save. Clock-based schedules use PT (America/Los_Angeles), including daylight saving changes.
• Check the next scheduled run in the agent’s history.

Deliver scheduled results to Slack

With a scheduled trigger enabled, turn on Slack delivery > Send output to Slack, choose channels, and Save.

Slack receives the scheduled run’s final assistant message, not its full tool transcript. Check delivery after the next scheduled run; manual Run now does not send scheduled Slack output.

Inspect history and pause recurring work

• Select a run on the agent’s page to read its conversation and status. Use Show more… for older runs.
• To pause future runs, turn off Settings > Scheduled trigger and Save. This keeps the agent available and does not stop a run already in progress.

Custom-agent owners and organization administrators can manage its settings. Other readers may see View settings. Built-in agents have different controls; these scheduling steps apply to custom Background Agents.

Troubleshooting

• If Run now is disabled, provide saved instructions or a run prompt and check whether a run is already starting.
• If saving is disabled, check the name, instructions, schedule validation, and whether you have changed a setting.
• If the schedule fails, inspect any error shown under Triggers and the run history before assuming the task found nothing.
• If Slack controls are disabled, enable the scheduled trigger first. If the channel is unavailable, check the workspace connection and channel access in Settings > Connections.

Next step

Add project Knowledge for durable context that the agent should reuse across investigations.