MLE-agent
MLE-agent is an LLM-powered assistant that helps machine learning engineers and researchers automate AI/ML tasks, from building baselines to participating in Kaggle competitions. It integrates with Arxiv and Papers with Code, offers smart debugging, and provides an interactive CLI chat for project enhancement.
✨ Key features
- Autonomous baseline creation from requirements
- End-to-end Kaggle competition participation
- Arxiv and Papers with Code integration
- Smart debugging with automatic code fixes
- File system and MLOps tool integration
- Interactive CLI chat and weekly report generation
🎯 Use cases
- Prototype an ML baseline from vague requirements
- Generate weekly work reports from GitHub or local git
- Participate in Kaggle competitions automatically
- Enhance existing ML projects via interactive chat
📦 Installation
🧰 Requirements: Python environment with pip or uv; requires an LLM API key (OpenAI, Anthropic, or Ollama) and optionally Kaggle credentials for Kaggle mode.
pip install -U mle-agentuv pip install -U mle-agentFrom source:
git clone https://github.com/MLSysOps/MLE-agent.git
cd MLE-agent
uv venv .venv
source .venv/bin/activate
pip install -e .
🚀 Usage
mle new
cd
mle start
For interactive chat:
mle chat
⚠️ Good to know
The project is under active development with some features still on the roadmap, such as Hugging Face integration and cloud deployment.
❓ FAQ
What LLMs are supported?
The README mentions support for Ollama LLama3, OpenAI GPTs, and Anthropic Claude 3.5 Sonnet.
How do I start a Kaggle competition?
Use mle kaggle in a project directory, or mle kaggle --auto with dataset and submission arguments for fully automated mode.
Can I generate a weekly report from my local git repository?
Yes, use mle report-local --email=<git email> --start-date=YYYY-MM-DD --end-date=YYYY-MM-DD.
Is there a web interface?
Yes, mle report starts a web application at http://localhost:3000/ for generating reports from GitHub.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.