CostGuards

cost guards focus on change, not targets while budgets track spend against a planned amount over time, cost guards watch for unexpected increases or decreases in cost and usage, regardless of whether a budget exists this makes them ideal for detecting anomalies, regressions, and runaway behavior as soon as they start why it matters not all cost changes are planned, and not all of them are bad some increases signal runaway usage some decreases indicate successful optimization others point to failures that quietly break production systems cost guard exists to catch unexpected changes in ai spend, both up and down, so you can investigate intent versus accident it gives you early visibility into behavior that falls outside normal patterns, before those changes turn into financial surprises or operational incidents this is how you detect meaningful change, not just track totals how to use it use cost guard when you want to monitor ai spend for abnormal behavior over time, rather than track it against a fixed target define what “unexpected” means cost guards are configured using change based conditions you can trigger a cost guard when spend increases by a specific dollar amount increases by a percentage decreases by a specific dollar amount decreases by a percentage each condition is evaluated over a defined time window, such as day over day, week over week, or month over month this lets you detect sudden shifts as well as slower moving trends scope guards to exactly what you care about cost guards can be narrowly scoped so alerts are meaningful instead of noisy for example, you can define guards that apply to a specific model a subset of services a particular business unit any combination of available dimensions this precision allows you to monitor critical workloads closely without being overwhelmed by irrelevant alerts detect both risk and regression increases in spend often indicate runaway usage, inefficient changes, or misconfiguration decreases in spend can be just as important a sharp drop in cost might mean teams successfully improved efficiency traffic stopped flowing due to a failure requests are being blocked, throttled, or failing silently cost guards help surface both scenarios so teams can confirm whether the change was intentional choose how alerts are delivered when a cost guard condition is met, you can choose how to be notified available options include email notifications sent to one or more recipients webhooks that trigger automated workflows email alerts ensure the right people are informed quickly webhooks allow cost guards to integrate with automation, such as throttling usage, triggering incident workflows, or applying protective controls use cost guards to automate protection because cost guards can trigger webhooks, they can be used as part of an automated response strategy for example a percentage increase in spend triggers throttling a spike in model usage initiates an automated investigation an unexpected drop triggers a system health check this turns cost guard from a passive alerting tool into an active protection mechanism track triggered alerts inside the platform cost guard alerts are visible in multiple places within the application this ensures that when an alert fires the notification is delivered the event is recorded you can review which guards triggered and when this creates traceability and accountability around cost related incidents when to use this page use cost guard when you want to catch unplanned changes in ai spend you need early warning signals for abnormal behavior you want to distinguish optimization from failure you need automated responses to cost anomalies cost guard helps you stay ahead of change, not just respond to it after the bill arrives