3 modalities
Tabular, graph, and time series
LLM-Driven Research Engineering
End-to-end anomaly detection from one command.
AD-AGENT turns natural-language requests into runnable anomaly detection pipelines across PyOD, PyGOD, and time-series libraries, with automated review and evaluation.
Supported by NSF POSE Phase II OpenAD (Award #2346158)
Multi-agent
Processor, Selector, Generator, Reviewer, Evaluator
Automation
Parse, select, codegen, test, and run
What It Does
AD-AGENT is a multi-agent framework for anomaly detection lifecycle automation: user intent parsing, model selection, documentation mining, executable code generation, synthetic-data validation, and real-data evaluation.
Core Features
Natural-language Interface
Write commands like: Run IForest on ./data/glass_train.mat and ./data/glass_test.mat.
Cross-library Support
Supports PyOD (tabular), PyGOD (graph), and TSLib/Darts (time-series).
Self-checking Pipeline
Generated code is reviewed on synthetic data before real execution.
Metric-driven Output
Reports AUROC/AUPRC and failure points for transparent evaluation.
Workflow
- 1Processor extracts algorithms, datasets, and params from user command.
- 2Selector infers data modality and selects AD library/tools.
- 3InfoMiner queries authoritative docs and model usage details.
- 4CodeGenerator creates runnable scripts and revises on errors.
- 5Reviewer tests on synthetic data; Evaluator runs on real data.
Quickstart
git clone git@github.com:USC-FORTIS/AD-AGENT.git
cd AD-AGENT
python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
python main.pyParallel mode: python main.py -p | Optimizer mode: python main.py -o
Citation
If this project helps your work, cite the paper:
@article{yang2025ad,
title={AD-AGENT: A Multi-agent Framework for End-to-end Anomaly Detection},
author={Yang, Tiankai and Liu, Junjun and ... and Zhao, Yue},
journal={arXiv preprint arXiv:2505.12594},
year={2025}
}Support
This project is supported by the U.S. National Science Foundation (NSF), TIP POSE program: NSF POSE: Phase II: OpenAD: An Integrated Open-Source Ecosystem for Anomaly Detection.