What It Does

Use Run Assistant Agent when you want an AI assistant to automatically triage a new alert. The agent analyzes the alert, then posts its findings as a threaded reply under the alert message — keeping all AI output tucked under the alert instead of adding channel noise.

You choose which agent runs: a built-in agent (RCA) or a reusable Assistant Agent you configured in Agent Studio with a custom prompt, integrations, and reference files.

<aside> ⚠️

This action must follow a Publish to Channel action. Run Assistant Agent is reply-only — it replies under the alert card that a Publish to Channel step posted earlier in the same automation. Add a Publish to Channel action above it, otherwise the automation cannot be saved.

</aside>

When To Use It

• Automatically kick off root-cause analysis the moment a high-severity alert fires
• Give responders a head start with likely causes and next steps before anyone opens the alert
• Run a custom, runbook-aware agent for a specific class of alerts (e.g. database issues)

Before You Start

• The automation must use the Alert Created trigger — Run Assistant Agent only runs for new alerts.
• The automation must include a Publish to Channel action ordered before this one.
• To use a customized agent, create it first in Agent Studio. Otherwise pick the built-in RCA agent.

How It Works

1. The alert-created automation fires and a Publish to Channel step posts the alert card.
2. Run Assistant Agent posts the agent's command as a reply under that card, the way a person would.
3. The selected agent analyzes the alert, using its instructions, allowed integrations, and any attachments.
4. The findings are posted as a threaded reply under the alert card. The card itself is not edited — everything the agent produces stays in the thread.

If no alert card was posted — because the Publish to Channel step was skipped by its conditions or failed — the action silently does nothing. There is nothing to reply under, and the user never saw an alert.

If the automation has Group burst alerts into one analysis turned on, the assistant does not run once per alert when several arrive together. The first alerts still run as above. The alert that brings the count to the threshold starts a group and waits; every alert this automation admits during the wait gets a short reply pointing at that first alert instead of its own analysis. When the wait ends, the agent runs once over the whole group and posts one analysis under the alert that started it. See Grouping burst alerts below.