AutoGluon Assistant (MLZero) is a multi-agent system designed for end-to-end multimodal machine learning automation. It empowers users to transform raw multimodal data into high-quality ML solutions with minimal human intervention, targeting researchers and practitioners seeking to streamline complex ML workflows. The system leverages LLMs to automate the entire ML lifecycle.

How It Works

MLZero employs a multi-agent architecture to automate multimodal ML/DL workflows. It processes raw data and generates ML solutions autonomously. The core algorithm incorporates a Node-Based Manager and Monte Carlo Tree Search (MCTS), drawing inspiration from ML-Master and AIDE, to enhance its decision-making and automation capabilities.

Quick Start & Requirements

• Installation: Linux-only. Install via pip: pip install uv && uv pip install git+https://github.com/autogluon/autogluon-assistant.git. Conda is a prerequisite.
• Prerequisites: Python 3.10-3.12. AWS credentials are required for the default Bedrock LLM provider; other providers (OpenAI, Anthropic, SageMaker) are supported. Docker is recommended for security. GPU support is available via Docker.
• Documentation: Tutorials are available for LLM providers, interfaces, configuration, and chat mode. Project page: https://project-mlzero.github.io/

Highlighted Details

• Automates end-to-end multimodal ML/DL workflows with zero human intervention.
• Supports multiple LLM providers including AWS Bedrock, OpenAI, Anthropic, and SageMaker.
• Offers diverse interfaces: CLI, Web UI, and MCP (Model Context Protocol).
• Features a new "Chat Mode" for interactive ML guidance without code execution.
• Incorporates advanced algorithms like Node-Based Manager and MCTS for enhanced performance.
• Accepted for presentation at NeurIPS 2025.

Maintenance & Community

Continuous Integration is managed via GitHub Actions. Specific community channels (e.g., Discord, Slack) or a public roadmap are not detailed in the README.

Licensing & Compatibility

The project is licensed under the Apache 2.0 license, which generally permits commercial use and integration into closed-source projects.

Limitations & Caveats

Currently supports Linux only. The execution of LLM-generated code necessitates caution, with Docker recommended for isolation. Installation from source may require managing complex dependencies.