An LLM-driven multi-agent framework that automates anomaly detection across tabular, graph, and time-series data.
Generate, execute, review, and evaluate anomaly detection pipelines using conversational prompts.
AD-AGENT is an open-source framework that converts natural language requests into executable anomaly detection code. It coordinates specialized LLM agents to select algorithms, fetch documentation, generate scripts, and evaluate performance. You should use it if you analyze operational data and want to automate detection workflows across tabular, graph, or time-series datasets.
The system accepts natural language instructions to run detection algorithms like Isolation Forest or DOMINANT on your raw datasets. It orchestrates separate agents for task planning, model selection, code generation, code review, and evaluation. Execution happens inside isolated Modal or Docker sandboxes to prevent unverified LLM code from running directly on your host machine.
It suits data analysts, technical project leaders, and operational engineers looking to prototype anomaly detection routines without manual script wiring. It is less suitable for non-technical teams needing a finished GUI or production-grade stability, as the project is still under active refactoring. It also requires an OpenAI API key and container infrastructure to run.
Converts text prompts into runnable Python scripts for anomaly detection algorithms.
Coordinates specialized agents to manage model selection, doc retrieval, code writing, code review, and evaluation.
Integrates with PyOD for tabular data, PyGOD for graph structures, and TSB-AD for time-series datasets.
Executes LLM-generated code inside local Docker containers or remote Modal environments for security.
Exposes individual orchestration stages as Python functions to embed directly into custom Jupyter notebooks or scripts.
AD-AGENT is free under the BSD-2-Clause license. Running it requires an active OpenAI API key, which incurs usage costs per token, alongside optional infrastructure costs if you use Modal remote sandboxes.
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.
AD-AGENT handles tabular, graph, and time-series data using integrations with PyOD, PyGOD, and TSB-AD libraries.
Yes, generated scripts run inside an isolated sandbox environment using either local Docker containers or remote Modal backends.
Yes, AD-AGENT relies on OpenAI models for agent reasoning and code generation, so an active OpenAI API key is required.
See Zekai first in Google