Skip to content

Alerts

Alerts are built in Data Processing → Alert Builder. Each alert (a trigger) runs a query on your data, checks a rule, and notifies the channels you connect.

%%{init: {"flowchart": {"curve": "linear"}}}%%
graph LR
    A["Query<br/>what to watch"] --> B["Rule<br/>deterministic or AI"]
    B -->|match| C["Slack"]
    B -->|match| D["Email"]
    B -->|match| E["Webhook"]

Create an alert

  1. Open Data Processing → Alert Builder and click Create new trigger.
  2. Enter a Trigger name and Trigger description.
  3. Configure Query: choose the endpoint, the columns and filters, and the time window to check.
  4. Configure Trigger Rule: pick a rule type (below) and set the condition.
  5. Configure Trigger Alerting Method: connect one or more channels and write the alert message.
  6. Save.

To make a variation of an existing alert, use Clone trigger.

Rule types

Type How it decides Good for
Deterministic Fixed conditions on the query result, including pattern (regex) rules Thresholds (error count, cost, latency) and sensitive-data checks
AI Triggered A model reads the query result and decides using a rule you write in plain language Fuzzy conditions such as "the agent promised a refund it can't give"

The rule library includes ready-made patterns, for example detecting email addresses, credit card numbers or US Social Security numbers in prompts or responses. Use these to catch sensitive data leaking through your AI features.

Channels

Channel Setup
Slack Click Connect to Slack, then choose the default channel
Email Add recipient email addresses
Webhooks Add one or more webhook URLs to call (for example your incident tool)

Tips

  • Start by sending a new alert only to yourself. Tune it for a few days, then add the team.
  • Put the useful context in the alert message (which agent, which endpoint, a dashboard link) so people can act without searching.
  • To run an alert as part of a bigger scheduled job, add it to a pipeline.