finds.dev← search

// the find

earthgecko/skyline

★ 594 · Python · NOASSERTION · updated Oct 2026

Real-time anomaly detection for time series metrics. 11 years of production anomaly detection evolution. Multi-algorithm ensembles. Pattern recognition using semi-supervised and unsupervised learning. Correlation analysis. Learns and gets better over time. No GPU required, can run on a VPS.

Skyline is a Python anomaly detection system for high-resolution time series, built as a set of specialized daemons (analyzer, mirage, boundary, ionosphere, luminosity, etc.) that ingest from Graphite, Prometheus, InfluxDB, or VictoriaMetrics. It's aimed at ops teams who need per-metric anomaly detection at scale without hand-tuning thresholds or running GPUs, and who are willing to operate a fairly heavy stack to get it.

Ionosphere's feedback loop is the standout feature — you confirm what wasn't actually anomalous, and it uses that per-metric to suppress future false positives, which is a real answer to the alert-fatigue problem most anomaly detectors ignore. The custom_algorithms directory has working implementations of isolation forest, LOF, PCA, matrix profile, spectral residual, Prophet, and MSTL already wired into the pipeline, not just one hardcoded method. Eleven years of running against real production metrics shows in the architecture — it's decomposed into single-purpose daemons (vista for remote fetch, luminosity for correlation, thunder for alerting) rather than one monolithic process.

Operating this is not light: Redis, MySQL, and a dozen separate daemons (bin/*.d) for what many teams want to be a single service — this is a platform, not a library you drop in. CI is still on Travis (.travis.yml) despite a push in October 2026, and the docs tree carries visible Python 2 / virtualenv cruft in the deprecated-docs section, suggesting parts of the operational tooling haven't kept pace with the core algorithm work. The README leads with a pitch for Anomify, the commercial managed version, which is a little awkward for an open source project's front page and makes you wonder how much future development attention the free version will keep getting.

View on GitHub → Homepage ↗

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →