Luminaire is an open-source Python library from Zillow that provides machine-learning-driven anomaly detection and forecasting for time series data. Available on GitHub with about 810 stars under the Apache-2.0 license and updated in 2026, it is designed as a largely hands-off way to monitor streams of measurements.
Monitoring time series at scale usually means hand-tuning thresholds for every metric. Luminaire instead offers several anomaly detection and forecasting methods that incorporate correlational and seasonal patterns, so the models adapt to the natural rhythms in the data. It includes a defined workflow for time series outlier detection and dedicated support for anomaly detection on high-frequency time series, where data arrives too fast for naive approaches. The package is published on PyPI and documented on its project site.
The library grew out of Zillow’s own need to watch large numbers of operational and business time series, where seasonal effects โ daily and weekly cycles โ make naive threshold alerts noisy and unreliable. Its models are built to learn those rhythms, along with correlations between related series, so deviations stand out more clearly. The documentation includes a step-by-step outlier-detection workflow, examples for high-frequency data such as minute-level streams, and references to blog posts from the development team explaining the methods in more depth.
Luminaire is free and open source under the Apache-2.0 license. There is no hosted tier or paid plan; you run the library inside your own data pipelines.
Luminaire fits data scientists, analytics engineers, and SRE or operations teams who monitor many time series โ product metrics, infrastructure signals, or business KPIs โ and want learned anomaly detection instead of static thresholds.