PyOD is the most comprehensive Python library for anomaly and outlier detection, covering more than sixty algorithms across tabular, time series, graph, text, image and audio data. Its latest generation adds an agentic orchestration engine so AI agents can run detection workflows end to end. PyOD has about 10,100 stars on GitHub and is actively developed in 2026.
Outlier detection matters everywhere — fraud, intrusion, manufacturing defects, data quality — but algorithms were scattered across papers and incompatible code. PyOD unifies them behind one scikit-learn-style API: classic methods like Isolation Forest, LOF, One-Class SVM and autoencoders sit beside modern deep and graph-based detectors, all with the same fit and decision calls, plus ensemble tools to combine them. Benchmarks against standard datasets (via the companion ADBench) back the implementations, and visualizations help explain flagged points. PyOD 3’s ADEngine orchestrates model selection and evaluation, with an agentic workflow layer that lets LLM agents propose, run and compare detectors — turning anomaly detection into a callable capability for AI systems as well as a library for analysts. PyOD’s design goal is that switching detectors should cost one line of code, keeping model comparison studies and production migrations cheap. It connects to the wider ecosystem through the SUOD acceleration framework for large datasets and the PyGOD sister library for graph anomaly detection. The library is maintained by the research group behind the field’s standard benchmark suite, and its implementations are frequently cited in anomaly detection papers — a sign they are trusted as reference code.
PyOD is completely free and open source under the BSD 2-Clause license, including for commercial products and services.
PyOD is for data scientists, security and fraud analysts, and AI engineers who need production-ready anomaly detection — from quick baselines to agent-driven monitoring pipelines.