Bruin AI Step 4 of 4

Build a Self-Healing Pipeline Agent

Guide overview
  1. Understand the Self-Healing Loop6 min
  2. Create the Context Layer12 min
  3. Set Up Agent Access12 min
  4. 4Run the Self-Healing Loop12 min
PreviousSet Up Agent AccessFinishBack to Bruin AI

Capture the failure

An alert starts an incident. It does not automatically give the agent permission to change production. In this example, an engineer asks the Bruin agent to investigate a failed self-heal-shop run from Slack.

A Bruin Slack notification shows a failed self-heal-shop pipeline run, followed by a request to investigate it.

What to do

  • Record the pipeline name and run ID from the alert.
  • Record the failed asset and the failing check or error message.
  • Ask the agent to diagnose the problem before it reruns anything.

Why the agent needs this: the run ID ties every log, pull request, and verification result to the same incident.

Sign up to our newsletter

Practical updates on open-source data pipelines, AI analysts, governance, and what we are shipping at Bruin.

The signup form is hosted by Brevo. Accept cookies to load it.