AI alert examples

High Token Usage from One Model

• Goal: Alert if token usage spikes from GPT-4-turbo in a short period.
• Filter: Model Name = Claude Instant
• Condition: Number of AI logs is greater than 100
  Within: the last 1 hour
• Metric: Total Tokens | sum
• Use Case: Prevent overuse or cost spikes from high token activity.

Stop Reason: max_tokens Reached

• Goal: Know if many completions are cut off by max_tokens.
• Filter: Stop Reason = max_tokens
• Condition: Number of AI logs is greater than 10
  Within: the last 6 hours
• Use Case: Adjust token limits or prompt design if outputs are consistently too long.

Unexpected Latency Spikes

• Goal: Alert if LLM response time becomes unacceptably high.
• Filter: Model Provider = Amazon
• Condition: Number of AI logs is greater than 20
  Within: the last 15 minutes
• Metric: Latency | avg >
• Use Case: Monitor latency SLAs or performance degradation.