Optional ReadonlyaccountToken-use alarms on the account-wide total across all model IDs.
Use cases: overall token-spend alerting without enumerating models
AWS: CloudWatch alarms on the undimensioned AWS/Bedrock InputTokenCount / OutputTokenCount
Validation: Optional; TokenThresholdsProps
Optional ReadonlydatapointsNumber of breaching periods within evaluationPeriods required to alarm (an "M of
N" alarm).
Use cases: ignoring a single burst while catching sustained token use
AWS: CloudWatch alarm datapointsToAlarm
Validation: Optional; Number; between 1 and evaluationPeriods
Optional ReadonlyevaluationNumber of periods over which token use is compared to the threshold.
Use cases: requiring sustained token use before alarming
AWS: CloudWatch alarm evaluation periods
Validation: Optional; Number; periodSeconds x evaluationPeriods at most 86400 for a period under 3600, at most 604800 otherwise
Optional ReadonlymodelsToken-use alarms per model, keyed by a name for the model. Each entry lists every model and inference-profile ID callers use for that model.
Use cases: per-model token-spend alerting
AWS: CloudWatch alarms on AWS/Bedrock token metrics by ModelId
Validation: Optional; NamedModelTokenAlarmProps
Optional ReadonlyperiodAlarm period in seconds. Each period's token count is compared to the threshold.
Use cases: matching the alarm window to the token budget's granularity
AWS: CloudWatch alarm period
Bedrock publishes token metrics at one-minute resolution, so the period is a whole number of minutes.
Validation: Optional; Number; multiple of 60; periodSeconds x evaluationPeriods at most 86400 for a period under 3600, at most 604800 otherwise
Alarm selection and evaluation timing shared by both notification targets of TokenUsageAlarmsProps.