pyod: a comprehensive anomaly detection library with agentic workflows for multi-modal data

What it solves

PyOD is a comprehensive Python library designed to detect outliers and anomalies in datasets. It provides a unified API for a wide variety of detection algorithms, allowing users to easily switch between different methods to find the most effective one for their specific data.

How it works

PyOD provides three layers of usage:

1. Classic API: A standard fit/predict interface for users who already know which specific detector they want to use.
2. ADEngine: An orchestration core that automatically chooses, compares, and assesses detectors for the user.
3. Agentic Investigation: An AI-driven layer where natural language requests are converted into workflows via the od-expert skill (for Claude Code/Codex) or an MCP server for other LLM-compatible agents.

Who it’s for

It is intended for data scientists, researchers, and AI engineers who need to perform anomaly detection across multiple data modalities, including tabular, time series, graph, text, image, and audio data.

Highlights

• Multi-Modal Support: Includes 61 detectors covering tabular, time series, graph, text, image, and audio data.
• Agentic Workflows: Integrates with LLM agents via MCP and the od-expert skill to drive investigations through conversation.
• Scale and Adoption: Over 46 million downloads and utilized by organizations like Walmart and the European Space Agency.
• Comprehensive Benchmarking: Backed by benchmarks such as ADBench, TSB-AD, and BOND.
• High Performance: Built on SUOD for parallel training and utilizes numba JIT for speedups.