signalguard-aiops

Time-series and metrics AIOps toolkit for anomaly detection and SLO-aware incident signals.

Python package Part of the SignalGuard observability suite

What it is

signalguard-aiops is a lightweight library for time-series anomaly detection: z-score, EMA, Isolation Forest, LOF, Prophet residuals, LSTM autoencoders, and opinionated AIOps recipes.

Core use cases

• Error-rate incident detection
• Latency SLO breach detection (p95/p99)
• Ensemble anomaly scoring across detectors
• Prometheus-integrated incident pipelines

Tech stack

Python · NumPy · Pandas · scikit-learn · Prophet · TensorFlow · Prometheus HTTP API adapter.

Repository

View the GitHub repository →

Quick start

Install in editable mode and run the included recipes on synthetic or Prometheus-backed metrics.

Install CLI 
git clone https://github.com/danial-amin/signalguard-aiops
cd signalguard-aiops
pip install -e .
Error-rate recipe Metrics 
from signalguard_aiops.metrics import TimeSeries
from signalguard_aiops.recipes import ErrorRateZScoreRecipe
from signalguard_aiops.incidents import IncidentScorer

ts = TimeSeries.from_lists(timestamps, values, name="error_rate")
recipe = ErrorRateZScoreRecipe(service="checkout")
incident = recipe.run(ts)
severity = IncidentScorer.simple_severity(incident)
Latency SLO recipe SLO 
from signalguard_aiops.recipes import LatencySLORecipe

recipe = LatencySLORecipe(
    service="payments",
    metric="latency_p95",
    slo_ms=300.0,
)
incident = recipe.run(ts)
Prometheus integration AIOps 
from signalguard_aiops.pipelines import PrometheusSeriesFetcher

fetcher = PrometheusSeriesFetcher("http://localhost:9090")
series = fetcher.fetch_range(
    query='rate(app_request_errors_total[5m])',
    start_ts=..., end_ts=..., step="30s"
)

Included detectors

ZScoreDetector · EMADetector · IsolationForestDetector · LOFDetector · ProphetResidualDetector · LSTMAutoencoderDetector

These detectors are exposed behind a common interface and can be composed into higher-level recipes for error-rate incidents, latency SLOs, and ensemble scoring across heterogeneous models.

AIOps recipes

Opinionated building blocks for production-style monitoring flows.

Examples:
• ErrorRateZScoreRecipe: simple, interpretable baseline.
• ErrorRateIForestRecipe: non-Gaussian error distributions.
• LatencySLORecipe: SLO-aware latency anomaly detection.
• EnsembleErrorRateRecipe: majority-vote ensemble combining multiple detectors.