Health and SRE Dashboard
Operational monitoring for your agent governance infrastructure. This is about Zabta's own performance — evaluation latency, error rates, and uptime — not your agent's behavior.
What it monitors
The health dashboard provides SRE-style operational monitoring for the governance layer that sits between your agents and their actions. When Zabta is slow or down, your agents are affected — either they wait for policy evaluation or they proceed ungoverned. The health dashboard helps you detect and respond to infrastructure issues before they impact agent operations.
Key metrics
Evaluation latency
Target: < 500ms at p95Time from evaluate() call to response. Tracked at p50, p95, and p99 percentiles. Latency above target may indicate cache misses or increased policy complexity.
Error rate
Target: < 0.1%Failed evaluations divided by total evaluations. Excludes policy denials (those are correct behavior). Tracks only infrastructure-level failures.
Policy cache hit rate
Target: > 90%Percentage of evaluations served from cache. Low hit rates increase latency and backend load. Cache misses spike after jurisdiction pack changes.
API uptime
Target: 99.9%Availability of the policy evaluation endpoint. Measured over rolling 30-day windows.
Evaluation throughput
Target: VariesEvaluations per minute across all agents. Helps with capacity planning and identifying usage spikes.
Alerting
The health dashboard surfaces issues before they impact agents:
- Latency alerts — triggered when p95 evaluation latency exceeds the configured threshold for 5+ consecutive minutes
- Error rate alerts — triggered when the error rate exceeds 1% over a 5-minute window
- Cache degradation — triggered when the cache hit rate drops below 70%, indicating potential configuration issues
- Throughput anomalies — triggered when evaluation volume deviates significantly from the rolling average, which may indicate agent misconfigurations or runaway loops
Related