pyod
A Python library for anomaly detection across tabular, time series, graph, text, image, and audio data. 60+ detectors, benchmark-backed ADEngine orchestration,
What is pyod?
A Python library for anomaly detection across tabular, time series, graph, text, image, and audio data. 60+ detectors, benchmark-backed ADEngine orchestration, and an agentic workflow for AI agents.
Overview of pyod
What it is
PyOD (Python Outlier Detection) is a library for anomaly detection in Python. It provides over 60 detectors that work on tabular, time series, graph, text, image, and audio data. Version 3 introduces ADEngine, a lifecycle orchestration tool, and an agentic workflow that allows AI agents to use the library through natural language. The library keeps its classic fit/predict API unchanged. PyOD has been downloaded over 46 million times and is used in both academic research and commercial products.
Key features
- Multi-modal support: 61 detectors covering tabular, time series, graph, text, image, and audio data under a single API
- Full lifecycle management: ADEngine handles detection from raw data to explained anomalies and next-step guidance in one call
- Agentic: The od-expert skill lets Claude Code and Codex drive ADEngine investigations; an MCP server exposes tools for other LLMs
- Benchmark-backed routing: Uses results from ADBench, TSB-AD, BOND, and NLP-ADBench to recommend detectors
- Three usage layers: Classic API for known detectors, ADEngine for automatic selection and comparison, and agentic investigation for natural-language-driven workflows
Use cases
Anomaly detection on structured and unstructured data, including tabular datasets, time series, graphs, text documents, images, and audio recordings. The library suits users who need to quickly evaluate multiple detectors, automate detection pipelines, or integrate anomaly detection into agent-based AI systems. ADEngine is designed for end-to-end analysis from raw data to interpretable findings, while the agentic layer supports interactive multi-turn investigations through natural conversation.
Getting started
pip install pyodFor the agentic skill:pyod install skill # Claude Code: user-globalpyod install skill --project # Codex: project-localFor the MCP server:pip install pyod[mcp]pyod mcp serveBasic usage example:pythonfrom pyod.models.iforest import IForestclf = IForest()clf.fit(X_train)y_train_scores = clf.decision_scores_y_test_scores = clf.decision_function(X_test)See the repository README for conda, source install, and troubleshooting details.
License and pricing
Licensed under BSD-2-Clause. Open source.
Based on the project's public README and repository metadata.
