LLM Anomaly
Detection
Stop finding out about LLM cost spikes from your invoice. Sentinelboard automatically detects unusual spending patterns and alerts you in real time - no thresholds to configure.
Free tier available · No credit card required
LLM costs are unpredictable by nature
A prompt regression causes your LLM token usage to spike 10x and you don't notice until the invoice arrives.
A retry loop burns through your entire monthly budget in a single afternoon.
Weekend traffic patterns look nothing like weekdays, so flat thresholds either miss real spikes or cry wolf constantly.
How anomaly detection works
Static spending thresholds don't work for LLM costs. Usage varies by day of the week, by feature adoption, and by the prompts your users send. A fixed limit either fires too often or misses real problems entirely.
Sentinelboard takes a different approach. It builds a statistical baseline from your actual spending history and flags days that deviate significantly from what's normal for you. The detection adapts as it collects more data - starting with simple median-based checks for new metrics and graduating to weekday-normalized z-score analysis once it has two weeks of history.
Detection runs both in real time (as LLM events arrive) and as a nightly aggregation. That means you catch runaway costs within minutes during the day, and get a clean summary of each completed day overnight. When an anomaly is detected, your metric status escalates to a warning and alerts fire through your configured channels.
Gets smarter with more data
The detection strategy automatically upgrades as your spending history grows. No manual tuning required.
Cold start
Compares daily cost against the median of historical days. Flags anything above 5x the median as anomalous.
Z-score
With enough history for statistics, uses z-score analysis. Past anomalies are excluded from the baseline so one spike doesn't inflate the mean.
Weekday-normalized
Accounts for day-of-week patterns. High-spend weekdays and quiet weekends each get their own baseline, reducing false positives.
Everything you need for LLM cost anomaly detection
Automatic, adaptive, and zero-config. Just connect your OpenRouter metric and detection starts working immediately.
Adaptive detection tiers
The system automatically selects the best detection strategy based on how much history you have - from cold-start median checks to weekday-normalized z-scores.
Real-time spend monitoring
Every LLM call triggers a check against your historical baseline. Anomalies are caught as they happen, not the next morning.
Weekday-aware baselines
With two weeks of data, detection adjusts for day-of-week patterns. High-spend Mondays won't trigger false alarms, and quiet-Sunday spikes won't slip through.
Multi-channel alerts
Get notified via email, Slack, Telegram, Discord, or webhooks the moment an anomaly is detected.
Zero configuration
Anomaly detection works out of the box. No thresholds to set, no rules to write. Just connect your OpenRouter metric and monitoring starts automatically.
Visual anomaly indicators
Anomalous days are highlighted directly on your spending chart with red indicators and detailed tooltips showing the anomaly score.
Stop worrying. Start shipping.
Join hundreds of indie developers who sleep better knowing their apps are monitored 24/7.
Free tier available · No credit card required