Job Outcome Alerts

dbt Cloud produces a pass or fail result each time a scheduled job completes. A Beam agent reads the outcome as it lands, applies the account's rule for how to respond, for example notifying the job owner the moment a production job fails, and updates a status record so the team can see history at a glance. Where the rule allows a single automatic retry for a known flaky step, the agent applies it. Repeated failures, or a failure in a job the rule doesn't cover, are left for an engineer to dig into directly.

Test Failure Alerts

Models in dbt Cloud carry data tests, checks like uniqueness or a value never being null, that execute alongside each job. A Beam agent reads which tests failed and how badly, matches that against the account's rule for severity, and notifies the model owner or a data steward when a test crosses the threshold that matters. A handful of warning-level failures on a known noisy test are usually just noted. A failure on a test tied to a customer-facing dashboard, or several tests failing together, gets escalated for a person to look into before anyone trusts the numbers downstream.

Source Freshness Checks

dbt Cloud can check whether a source table was loaded recently enough to trust, based on a warn and error threshold set for it. A Beam agent reads the freshness result for each source, applies the account's rule for what a stale source means, and notifies whoever owns the pipeline feeding it once a source crosses the warn line. Sources missing a freshness check entirely, or ones where the underlying load job itself has stopped, are flagged for an engineer to chase rather than left to age quietly. This keeps stale data from reaching a report before someone notices.