When something is judged abnormal, an investigation begins. AI reads the schema and follows the related data with read-only queries to test its hypotheses. The notification you receive includes the cause, impact, evidence, and recommended action.
Orders at 9 p.m. went above the usual range. Checking how this differs from the nightly peak, and which values moved with it.
Cycle · hour
Even after taking out the usual 9 p.m. peak, tonight's orders are about 40% above normal. They started rising one hour after the number of coupons sent went up.
Alerts come in, but only one person can investigate the cause
Anomalies at night or on weekends go uninvestigated until morning
The investigation isn't recorded, so the same problem starts from scratch next time
AI builds and runs queries, then looks at the results to decide its next step. From what it learns, it narrows down where the cause lies.
If the investigation shows a false alarm, the incident is closed without a notification. Only real anomalies reach you.
They arrive with the cause, impact, evidence, and recommended action. What you need to decide is business knowledge, not SQL.
AI's messages, the queries it ran and their results, and the charts it presented are kept in order. While it runs, you can watch it live on screen.
AI has no shell and no file access. It can only use a fixed set of tools. Connection strings are never given to AI; connections are referenced by ID.
Reads table and column structure to decide where to look.
Runs only single SELECT / EXPLAIN / SHOW statements over a read-only connection.
Checks hourly and weekly cycles, and how other values move together, by calculation, then charts them in the record.
Pulls up what happened before on the same connection or monitor, and how it was concluded.
Decides whether it's a real anomaly or a false alarm, and records the cause, impact, evidence, and recommended action.
Saves facts learned during the investigation to the project's memory, for use in later judgments and investigations.
All you need is a database connection string or an account connection. Nothing to install on your servers.
An anomaly is detected
When a judgment finds something abnormal, an incident opens and an investigation begins.
AI examines the real data
It reads the schema, runs queries, and narrows down the cause from the results.
The conclusion arrives
If it's a real anomaly, you're notified with the cause and recommended action. If anything else concerns you, you can request further investigation from the same screen.
E-commerce
Trace a spike in payment failures
AI checks failure rates by payment method and pinpoints that failures are concentrated on one card issuer. It also confirms that order volume itself remains at the usual level before notifying you.
False alarm
Close a temporary blip on its own
If investigating a jump in values shows it's a plausible change for the business, the incident is closed without a notification. The evidence it found stays on record.
Anomaly detection
AI judges whether something is abnormal based on value trends, past alerts, and your project's memory.
Approval-gated actions
AI proposes only actions you've registered in advance, and runs them when a person approves.
Notifications & integrations
Deliver notifications and actions to 10 destinations, including Slack, Microsoft Teams, Google Chat, LINE WORKS, and kintone.
Leave the investigating
to AI, too.
What reaches you is a finished conclusion, and the whole process is kept on record.