Multi-agent framework that converts natural language requests into isolated, executable anomaly detection pipelines.
Generate, validate, and run PyOD, PyGOD, and time-series anomaly detection scripts from plain English prompts.
AD-AGENT is an open-source multi-agent framework designed to automate end-to-end anomaly detection across tabular, graph, and time-series data. You describe your detection task in plain text, and specialized LLM agents choose models, write Python code, review it, and execute it inside a secure sandbox. It suits technical teams and data professionals building automated diagnostic workflows.
The framework orchestrates specialized agents responsible for model selection, documentation retrieval, code generation, review, and evaluation. It generates execution scripts targeting PyOD, PyGOD, and TSB_AD libraries based on your dataset and prompt. Generated code runs inside isolated local Docker containers or remote Modal sandboxes to protect your environment.
This tool is built for data scientists, developers, and project engineers who need to run anomaly detection models without manually authoring boilerplate script setups. It is not intended for non-technical users, as it requires managing Python environments, API keys, and containerized runtime environments. Teams needing a graphical interface will need to use the Python API or CLI instead.
Converts plain English prompts directly into executable anomaly detection Python scripts.
Generates code covering PyOD for tabular data, PyGOD for graph structures, and TSB_AD for time-series datasets.
Executes untrusted generated code safely inside local Docker containers or remote Modal instances.
Exposes core pipeline stages so you can embed selection and generation steps directly inside custom scripts or notebooks.
AD-AGENT is free and open-source under the BSD-2-Clause license. Running the framework requires an OpenAI API key, so your operating costs scale with your LLM token usage. Additional infrastructure expenses may apply if you opt for remote Modal cloud sandboxes over local Docker execution.
877 open remote AI-training roles, updated November 16, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
See Zekai first in Google