What are AI alarm suggestions and how do I use them?

The platform can analyze your meter data and suggest alarm rules automatically. These AI suggestions appear when the system detects patterns or anomalies that could benefit from monitoring.

How AI suggestions work

1. The system looks at meters that have trained pattern models (built from historical data)
2. For each meter, it analyzes consumption patterns — hourly variations, daily cycles, seasonal trends, and statistical outliers
3. It generates suggested alarm rules with a confidence score (0–1) and a rationale explaining why this alarm would be useful
4. Suggestions may include potential savings estimates if the anomaly is caught early

Types of suggestions you might see

• Threshold suggestions: "Your electricity meter shows consistent peaks above 450 kWh between 9–11 AM. Setting a threshold at 520 kWh would catch unusual spikes." (confidence: 0.85)
• Pattern suggestions: "Weekend usage on this meter is normally near zero but has spiked 3 times in the last month. A pattern alarm would detect future after-hours usage." (confidence: 0.78)

How to use AI suggestions

1. Go to Analytics → Alarms — the dashboard strip shows the count of available suggestions (purple badge)
2. Review each suggestion: the system shows the suggested name, type, configuration, rationale, and confidence
3. To accept a suggestion, click "Accept" — this creates a real alarm rule from the suggestion, marked with ai_generated = true
4. The new rule is immediately active and will be evaluated by the alarm worker

Accepting a suggestion does not lock you in. After accepting, the alarm rule behaves like any other rule — you can edit the threshold, change the severity, adjust the cooldown, or disable it entirely.

When AI suggestions appear: They are generated on-demand when you visit the Alarms page. The system recalculates suggestions each time based on current pattern models. If no pattern models exist for any meters, no suggestions will appear.

Where to see them: Analytics → Alarms → look for the purple "suggestions" count in the dashboard strip.