AIDE
AIDE ML is an open-source Python package implementing the AIDE algorithm, an LLM-driven agent that autonomously writes, evaluates, and improves machine-learning code via tree search. It solves the problem of automating ML pipeline development from a dataset and a natural-language goal, targeting researchers and ML practitioners who want to replicate the paper or prototype solutions.
✨ Key features
- Natural-language task specification: describe goal and metric in plain English.
- Iterative agentic tree search: code scripts as nodes, patches as children, metric feedback guides search.
- Model-neutral: supports OpenAI, Anthropic, Gemini, or any local LLM with OpenAI API.
- HTML visualizer to inspect solution tree and code.
- Streamlit web UI for interactive prototyping.
- CLI and Python API for flexible experimentation.
🎯 Use cases
- Automatically build a high-performance ML pipeline for a given dataset.
- Replicate the AIDE paper results or test new search heuristics.
- Prototype ML solutions interactively via the Streamlit UI.
- Use as a benchmark for evaluating ML agents (as in MLE-bench).
📦 Installation
🧰 Requirements: Python environment with pip; requires an LLM API key (e.g., OPENAI_API_KEY) or a local LLM endpoint; optional Docker for containerized runs.
pip install -U aideml
For web UI or development:
git clone https://github.com/WecoAI/aideml.git
cd aideml
pip install -e . # adds streamlit
🚀 Usage
export OPENAI_API_KEY=<your-key>
aide data_dir="example_tasks/house_prices" \
goal="Predict the sales price for each house" \
eval="RMSE between log-prices"
After run, find logs/<timestamp>/best_solution.py and tree_plot.html.
⚠️ Good to know
The README notes that using fully local models may result in some performance drop; otherwise, no explicit limitations are stated.
❓ FAQ
What is the AIDE algorithm?
AIDE is a tree-search agent that autonomously drafts, debugs, and benchmarks code until a user-defined metric is maximized or minimized, as described in the paper.
Which LLMs can I use?
The tool is model-neutral and supports OpenAI, Anthropic, Gemini, or any local LLM that speaks the OpenAI API, such as Ollama.
How do I run the web UI?
Clone the repo, install with pip install -e ., then run streamlit run app.py from the aide/webui directory.
What outputs does a run produce?
It produces best_solution.py with the best code found and tree_plot.html to inspect the solution tree.
📊 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.