Controls Overview

Define custom rules to monitor KPIs, detect drift, identify anomalies, and configure alert channels.

What are Controls?

Controls are configurable rules that monitor your AI agents and trigger alerts when thresholds are breached. TuringPulse provides four types of controls:

📊

KPI Rules

Track metrics and alert on threshold breaches

📉

Drift Detection

Detect statistical changes in agent behavior

⚠️

Anomaly Rules

Identify outliers and unusual patterns

🔔

Alert Channels

Configure where alerts are sent

Creating Rules via UI

Navigate to Controls in the sidebar to create and manage rules:

Go to Controls → Thresholds (for KPIs)
Click Create Rule
Select the workflow(s) to apply the rule to
Configure the metric, threshold, and alert settings
Save and enable the rule

Creating Rules via SDK

You can also define rules directly in your code:

from turingpulse import instrument, KPIConfig

@instrument(
    agent_id="my-agent",
    kpis=[
        # Latency threshold
        KPIConfig(
            kpi_id="latency_ms",
            use_duration=True,
            alert_threshold=5000,
            comparator="gt",
            auto_create_incident=True,
        ),
        # Cost threshold
        KPIConfig(
            kpi_id="cost_usd",
            value=lambda ctx: ctx.metadata.get("cost", 0),
            alert_threshold=1.0,
            comparator="gt",
        ),
    ]
)
def my_agent(query: str):
    return process(query)

Rule Configuration Options

Common Options

Option	Description
`name`	Human-readable name for the rule
`workflow_id`	Apply to specific workflow, or "*" for all
`enabled`	Enable/disable the rule
`severity`	warning, critical
`auto_create_incident`	Automatically create incident on breach
`alert_channels`	List of channels to notify

KPI-Specific Options

Option	Description
`metric`	Metric to track (latency, tokens, cost, custom)
`threshold`	Numeric threshold value
`comparator`	gt, lt, gte, lte, eq
`window`	Time window for aggregation (e.g., "5m", "1h")

Drift-Specific Options

Option	Description
`metric`	Metric to monitor for drift
`baseline_window`	Historical window for baseline (e.g., "7d")
`detection_window`	Recent window to compare (e.g., "1h")
`sensitivity`	low, medium, high

Anomaly-Specific Options

Option	Description
`metric`	Metric to monitor for anomalies
`method`	zscore, iqr, isolation_forest
`threshold`	Anomaly score threshold
`min_samples`	Minimum samples before detection

Alert Channels

Configure where alerts are sent when rules are triggered:

Email - Send to individual or group emails
Slack - Post to Slack channels via webhook
PagerDuty - Create incidents in PagerDuty
Webhook - Send to any HTTP endpoint

💡

Best Practice

Create separate alert channels for different severity levels. Route critical alerts to PagerDuty and warnings to Slack.

Next Steps

KPI Rules - Detailed KPI configuration
Drift Detection - Set up drift monitoring
Anomaly Rules - Configure anomaly detection
Alert Channels - Set up notifications