For the complete documentation index, see llms.txt. This page is also available as Markdown.

Reactive Rules

Overview

Reactive Rules allow you to continuously monitor metrics and automatically execute configured actions when specified conditions are met.

A reactive rule evaluates data in the background against a defined condition. When the condition is satisfied, BlueRock performs the configured action, such as generating an in-app alert or creating an operational log entry.

Reactive Rules support fixed thresholds and adaptive statistical models. Fixed thresholds use explicit limits, while statistical models help identify changes or anomalies relative to observed activity.

For example, you can create a rule that triggers when LLM spend exceeds $50 per hour.

Note: Advisor and AI analysis are disabled in this release.


Detection Methods

Reactive Rules support the following detection methods:

Detection method

Description

Fixed thresholds

Trigger when a metric crosses a specified value.

Z-score anomaly detection

Identify values that differ significantly from an established baseline.

Top-K shifts

Detect changes in the entities appearing among the top-ranked results.

CUSUM changepoints

Detect sustained changes in the behavior of a metric.

Predictive forecasting

Identify conditions based on forecasted values and established activity patterns.

Statistical models use observed activity to adapt detection to normal operating patterns.


Create a Reactive Rule

You can create a reactive rule by defining the scope, selecting a metric and condition, and configuring the action to execute when the condition is met.

Procedure

  1. In the left navigation sidebar, select Reactive Rules.

  2. Select + New Rule in the upper-right corner.

  1. Configure the rule using the settings described in the following sections.


Configure the Rule

1. Name

Enter a descriptive name for the rule.

For example:

High active session count

2. GFL Expression

Define the rule scope using a Graph Filter Language (GFL) expression. The expression specifies the nodes or edges to evaluate.

For example:

ms.state == 'active'

The expression must be valid for the rule to be created.

3. Configure the Condition

Specify the metric and condition that BlueRock evaluates.

Setting

Description

Signal

Select the signal used for evaluation, such as Instant.

Metric

Select the metric to measure, such as count.

Entity type

Select the entity type to include in the evaluation.

Comparison operator

Select the operator used to compare the observed value against the threshold.

Threshold

Enter the value against which the observed metric is compared.

The available comparison operators shown in the configuration include:

  • >

  • >=

  • <

  • <=

  • =

For example, a condition can use count > 0.

4. Statistical Model (Optional)

Select a statistical model if you want to apply adaptive detection to the condition.

Available options shown in the UI include:

  • None

  • Z-score

  • Top-K shift

5. Configure the Action

Specify what BlueRock should do when the condition is met.

Setting

Description

Action Type

Select the action to execute, such as ALERT or LOG.

Severity

Select the severity when applicable. The source content identifies Info, Warning, and Critical.

Message

Enter an optional notification or log summary.

6. Configure Cooldown

Specify the minimum time that must elapse before the rule can trigger again.

For example:

  • 5m

  • 1h

A cooldown helps prevent repeated triggers while the underlying condition remains active.

7. Configure Scope and Personas

Configure the rule's visibility and optional classification.

Setting

Description

Scope

Specify the rule's visibility, such as User, Team, or Org.

Personas

Optionally assign categories such as cost, security, or reliability.

Enabled

Enable or disable the rule.

  1. Review the configuration and select Create.

BlueRock saves the rule. When enabled, the rule can evaluate its configured condition and execute the specified action when the condition is met.


View Reactive Rules

The Reactive Rules page displays the rules configured by the user and BlueRock pre-provisioned rules.

The rules table includes the following information:

Column

Description

Name

Name of the rule.

Expression

GFL expression configured for the rule.

Condition

Metric condition and threshold.

Action

Configured action, such as ALERT.

Personas

Personas assigned to the rule, if any.

Enabled

Indicates whether the rule is enabled.

Last triggered

Time when the rule last triggered.

Triggers

Number of triggers recorded for the rule.

The page also provides access to BlueRock pre-provisioned rules.


Edit a Reactive Rule

You can update the configuration of an existing reactive rule.

  1. Locate the rule in the rules table.

  2. Select the Edit icon in the row.

  3. Update the required settings.

  4. Select Save to apply the changes.


Duplicate a Reactive Rule

You can create a copy of an existing rule and use it as the starting point for another configuration.

  1. Locate the rule in the rules table.

  2. Select the Duplicate icon in the row.

  3. In the Duplicate Rule dialog, enter a name for the new rule.

  4. Review and update the expression, condition, action, and other settings as required.

  5. Select Create.


Evaluate a Reactive Rule

You can evaluate a reactive rule against current data to check its condition.

  1. Locate the rule in the rules table.

  2. Select Evaluate now using the rule's action icon.

  3. Review the evaluation result.

The UI displays an evaluation message indicating whether the condition was met and the observed value. For example, it may report that the condition was not met and show an observed value of 0.

An evaluation checks the rule against current data. It does not replace the rule's ongoing monitoring behavior.


How Reactive Rules Work

The reactive rule workflow is:

Define scope → Select metric → Define condition → Configure action → Set cooldown → Enable rule

Once enabled, BlueRock evaluates the configured condition. When the condition is met, the configured action executes.


Last updated