AI Analysis
An AI analysis is a short written explanation of how a sensor is doing right now. It reads the facts HydroNode has already checked and puts them into plain words, with findings and recommendations.
What the AI does and does not do
Scores, severities and incidents are calculated by HydroNode itself, with fixed rules, before the AI sees anything. The AI only explains that result. It may not contradict it, invent values, thresholds or causes, or declare something safe that a rule flagged.
What an analysis contains
| Part | What it is |
|---|---|
| Summary | One or two sentences on the overall state, naming only what stands out. |
| Findings | Up to 6. Each is marked Supported when a confirmed incident backs it, or Possible when it is an interpretation or a suggested cause. A suspected sensor fault is always Possible. |
| Recommendations | Up to 3 concrete next steps, chosen from a fixed list of allowed actions. |
| Limits | What the analysis could not judge, for example a missing unit, too little data or a battery without a configured type. |
You find the latest analysis in the anomaly panel and in the analysis section of the sensor page. Each analysis is a frozen snapshot: it shows the data it was based on, even after new readings arrive. Past analyses stay in the history for 90 days.
Three ways an analysis starts
| Trigger | When |
|---|---|
| Automatic | On a fixed schedule, if the sensor sent new readings since the last analysis. A silent sensor is not analysed. |
| Incident | When an important incident opens. HydroNode waits 60 seconds so related incidents end up in the same analysis. |
| Manual | Analyze in the anomaly panel or Refresh analysis on the sensor page. Owner only. |
Pressing the button before the minimum gap is over tells you when the next analysis is possible. If nothing has changed since the last analysis, you get that one back immediately instead of a new one.
Limits of a standard account
| Automatic analysis | … |
| Manual analyses | … |
| Same sensor | … |
| Per account | … |
What is sent to the model
The analysis runs on Amazon Bedrock in Frankfurt (EU). Your data is not used to train models.