// the find
chaos-genius/chaos_genius
ML powered analytics engine for outlier detection and root cause analysis.
Chaos Genius is a self-hosted anomaly detection and root cause analysis tool for business/data metrics. You point it at a data source, define KPIs and dimensions, and it runs time-series models to flag outliers and drill down into which dimension combinations drove a change. It is explicitly archived and no longer maintained.
Multiple anomaly models out of the box — Prophet, EWMA, EWSTD, Neural Prophet, Greykite — so you can pick what fits your time series rather than being stuck with one. The DeepDrills engine uses an A*-like search to avoid combinatorial explosion when scanning high-cardinality dimension combinations, which is the right approach to a genuinely hard problem. Connector coverage is solid for a project this size: Postgres, MySQL, BigQuery, Snowflake, Redshift, Clickhouse, Databricks, Athena, Druid all present. Docker-compose setup is real and functional, not aspirational.
The repo is archived — the README says so in the first section, and the last push was September 2024. Adopting this means owning the maintenance yourself from day one. Several headline features (seasonality detection, automated RCA, forecasting, what-if analysis) are listed as roadmap items that will never ship. The operational footprint is heavy for what it does: Flask app, Celery workers, a separate integration server, Redis, and a frontend container — that's five moving parts before you've written any queries. No RBAC or multi-tenancy, so if more than one team needs isolated KPI views, you're either running separate instances or patching it yourself.