Mastra Platform adds observability alerts: get notified when agent behavior crosses your limits, before the bill or a user does
Synopsis
Mastra Platform now offers observability alerts that notify users when model costs exceed limits, error rates climb, or eval scores drop, sending messages to Slack, Discord, Microsoft Teams, email, webhooks, PagerDuty, or incident.io. Previously users had to watch the Metrics page themselves or export traces and build alerts in tools such as Datadog or Grafana; now alerts are created inside the platform, with thresholds checked across all matching traffic or separately per agent, tool, model, or user.
Interpretation
Mastra Platform now supports observability alerts that fire when an agent's behavior falls outside the limits a user sets, even while the service is running normally. The platform already had incident alerts for failed deploys or crashed services; observability alerts cover the case where the service is healthy but behavior is out of bounds. The triggers the platform lists include error rate, number of errors, total count of matching traces or spans, latency, time to first token, time to first chunk, token usage by category, spend, and the average score from a chosen scorer.
Alert rules are created directly in the platform, where a user picks a template or starts blank to write a query and condition, sets a threshold, and chooses where notifications go. Previously a user had to watch the Metrics page or export traces and build an alert in a tool like Datadog or Grafana. A threshold can be checked across all matching traffic or separately for each agent, tool, model, or user; a Backtest chart shows when the rule would have opened and closed on recent traffic, and the platform checks enabled rules every five minutes.
Perspective
The feature targets projects that already send application traces to Mastra Platform; alerts are created in the platform dashboard, and notifications can go to Slack, Discord, Microsoft Teams, email, webhooks, PagerDuty, or incident.io.
The material states that the platform checks enabled rules every five minutes, so alert timeliness follows that check interval; it gives no alert accuracy or false-positive figures.
