OpenTelemetry Observability¶

airiskguard ships a built-in TelemetryExporter that emits traces and metrics for every evaluation — no extra code required beyond wiring up your OTel providers.

Install¶

pip install "airiskguard[opentelemetry]"

Quick start¶

Configure your OTel providers before creating RiskGuard, then pass a TelemetryExporter:

from opentelemetry import trace, metrics
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import BatchSpanProcessor
from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter
from opentelemetry.sdk.metrics import MeterProvider

# Traces → OTLP (Jaeger, Grafana Tempo, Datadog, …)
tracer_provider = TracerProvider()
tracer_provider.add_span_processor(BatchSpanProcessor(OTLPSpanExporter()))
trace.set_tracer_provider(tracer_provider)

metrics.set_meter_provider(MeterProvider())

from airiskguard import RiskGuard
from airiskguard.telemetry import TelemetryExporter

guard = RiskGuard(
    telemetry=TelemetryExporter(service_name="my-ai-app"),
)

Every guard.evaluate() call now emits a span and increments metrics automatically.

Emitted signals¶

Traces¶

Span status is set to ERROR when the evaluation is blocked, OK otherwise.

Metrics¶

All signals carry model_id, overall_risk, blocked, and service.name attributes.

Configuration options¶

TelemetryExporter(
    service_name="my-ai-app",       # added to all spans/metrics
    record_input_hash=True,         # include SHA-256 of input as span attribute
    record_checker_spans=False,     # emit child span per checker (adds overhead)
)

Graceful no-op¶

If opentelemetry-api is not installed, TelemetryExporter silently no-ops — no errors, no performance impact. This means you can ship the same code to environments with and without OTel configured.

exporter = TelemetryExporter()
print(exporter.enabled)  # False if opentelemetry-api not installed

Grafana dashboard¶

A sample Grafana dashboard JSON is available in examples/grafana_dashboard.json (coming soon). Key panels to build:

• Blocked rate — rate(airiskguard_evaluations_blocked_total[5m]) / rate(airiskguard_evaluations_total[5m])
• Risk score p95 — histogram quantile on airiskguard_risk_score
• Top flagging checkers — topk(5, airiskguard_checker_flagged_total)

Backends¶

TelemetryExporter works with any OTel-compatible backend: