Policies: Automated Governance on SUPERWISE AMP
Define what to monitor, when to alert, and how to respond. Part of SUPERWISE AMP—the control plane for governing AI systems.
Manual Monitoring
- Checking dashboards manually—if you remember
- Arbitrary thresholds that don't adapt
- Alerts that wake you up for nothing
- No historical context when issues occur
- Different rules for different models? Good luck.
- Missed hallucination spikes
Policy-Driven Governance
- Automated evaluation on your schedule
- Dynamic thresholds that learn from history
- Actionable alerts to the right channels
- Full incident context with every alert
- Consistent policies across all models
- Catch failures before users notice. By the time you notice, it’s too late.
Policy Types
Choose the right threshold strategy for your use case.
Static Threshold
Alert when metrics cross fixed boundaries. Perfect for SLAs and compliance requirements.
Moving Average
Detect gradual model degradation automatically. Automatically adapts to normal variance.
Distribution Comparison
Detect data drift and behavior shifts in LLM outputs
Scheduled Evaluation
Run policy checks on your schedule—hourly, daily, or custom cron expressions.
Create Policies in Code or UI
Set fixed boundaries for metrics. When values cross the threshold, alerts fire immediately. Perfect for SLAs and compliance requirements.
Dynamic thresholds calculated from historical data. Define deviation tolerance and window size—SUPERWISE handles the statistics.
Route alerts to the right team. Slack for warnings, PagerDuty for critical. Different channels for different severities.
import superwise as sw
from superwise_api.models.policy import (
StaticThresholdSettings,
MovingAverageThresholdSettings,
DataConfigStatistics,
TimeRangeConfig
)
# Create a static threshold policy
sw.policy.create(
name="accuracy-sla",
dataset_id=dataset.id,
data_config=DataConfigStatistics(
query={"measures": [f"{ds.id}.avgAccuracy"]},
time_range_config=TimeRangeConfig(
field_name="timestamp",
unit="HOUR",
value=1
)
),
cron_expression="*/15 * * * *", # Every 15 min
threshold_settings=StaticThresholdSettings(
condition_below_value=0.95, # Alert if < 95%
),
alert_on_status="HEALTHY_TO_UNHEALTHY",
destination_ids=[slack_channel_id]
)Clear Status at a Glance
Every policy has a status. Every status has meaning.
All thresholds within bounds
Threshold violation detected
Awaiting first evaluation
No data available for evaluation
Alert Where You Work
Send alerts to the channels your team already uses.
From Policy to Incident Resolution
1. Define Policy
Set metrics, thresholds, and evaluation schedule
2. SUPERWISE Evaluates
Automated checks run on your schedule
3. Violation Detected
Threshold crossed? Status changes to unhealthy
4. Alert + Incident
Team notified automatically with full context for investigation
Policies in Action
Watch how SUPERWISE policies automate governance and see how policies detect and respond to issues automatically.
Related Platform Capabilities
As part of SUPERWISE AMP, policies work together with other capabilities
Stop Manual Monitoring
Define once. Monitor forever. Get alerted when it matters.
Related Content
Continue your journey: Explore related solutions, use cases, and resources to find your next step.