LLM‑driven agent that writes, evaluates & improves machine‑learning code.

Use in Production? Try Weco →

AIDE ML is the open‑source “reference build” of the AIDE algorithm, a tree‑search agent that autonomously drafts, debugs and benchmarks code until a user‑defined metric is maximised (or minimised). It ships as a research‑friendly Python package with batteries‑included utilities (CLI, visualisation, config presets) so that academics and engineer‑researchers can replicate the paper, test new ideas, or prototyping ML pipelines.

• Agent‑architecture researchers – swap in new search heuristics, evaluators or LLM back‑ends.
• ML Practitioners – quickly build a high performance ML pipelines given a dataset.

• Natural‑language task specification Point the agent at a dataset and describe goal + metric in plain English. No YAML grids or bespoke wrappers. aide data_dir=… goal="Predict churn" eval="AUROC"
• Iterative agentic tree search Each python script becomes a node in a solution tree; LLM‑generated patches spawn children; metric feedback prunes and guides the search. OpenAI’s MLE‑Bench (75 Kaggle comps) found the tree‑search of AIDE wins 4 × more medals than the best linear agent (OpenHands).

Know another public project that cites or forks AIDE?
Open a PR and add it to the table!

# 1  Install
pip install -U aideml

# 2  Set an LLM key
export OPENAI_API_KEY=<your‑key>  # https://platform.openai.com/api-keys

# 3  Run an optimisation
aide data_dir="example_tasks/house_prices" \
     goal="Predict the sales price for each house" \
     eval="RMSE between log‑prices"

After the run finishes you’ll find:

• logs/<id>/best_solution.py – best code found
• logs/<id>/tree_plot.html – click to inspect the solution tree

pip install -U aideml   # adds streamlit
cd aide/webui
streamlit run app.py

Use the sidebar to paste your API key, upload data, set Goal & Metric, then press Run AIDE.

The UI shows live logs, the solution tree, and the best code.

# Choose a different coding model and run 50 steps
aide agent.code.model="claude-4-sonnet" \
     agent.steps=50 \
     data_dir=… goal=… eval=…

Common flags

import aide
import logging

def main():
    logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
    aide_logger = logging.getLogger("aide")
    aide_logger.setLevel(logging.INFO)
    print("Starting experiment...")
    exp = aide.Experiment(
        data_dir="example_tasks/bitcoin_price",  # replace this with your own directory
        goal="Build a time series forecasting model for bitcoin close price.",  # replace with your own goal description
        eval="RMSLE"  # replace with your own evaluation metric
    )

    best_solution = exp.run(steps=2)

    print(f"Best solution has validation metric: {best_solution.valid_metric}")
    print(f"Best solution code: {best_solution.code}")
    print("Experiment finished.")

if __name__ == '__main__':
    main()

Place a .env file in the project root so AIDE can load defaults before any requests:

OPENAI_API_KEY=your_key
OPENAI_BASE_URL=https://your-resource.openai.azure.com/openai/v1/

OPENAI_API_KEY is read from AZURE_OPENAI_API_KEY as well if unset. Quotes around OPENAI_BASE_URL from .env are stripped automatically.

When OPENAI_BASE_URL is set, the OpenAI backend uses Chat Completions (Azure-compatible gateways), not the hosted Responses API. Use your deployed model identifier for agent.code.model, agent.feedback.model, etc. (for example gpt-5.2).

export OPENAI_BASE_URL="http://localhost:11434/v1"
aide agent.code.model="qwen2.5" data_dir=… goal=… eval=…

export OPENAI_BASE_URL="http://localhost:11434/v1"
aide agent.code.model="qwen2.5" agent.feedback.model="qwen2.5" data_dir=… goal=… eval=…

Tip: Expect some performance drop with fully local models.

docker build -t aide .
docker run -it --rm \
  -v "${LOGS_DIR:-$(pwd)/logs}:/app/logs" \
  -v "${WORKSPACE_BASE:-$(pwd)/workspaces}:/app/workspaces" \
  -v "$(pwd)/aide/example_tasks:/app/data" \
  -e OPENAI_API_KEY="your-actual-api-key" \
  aide data_dir=/app/data/house_prices goal="Predict price" eval="RMSE"

git clone https://github.com/WecoAI/aideml.git
cd aideml && pip install -e .

The default path is offline: collect heuristic AIDE runs, build a decision dataset with subtree-best rewards, SFT-imitate the heuristic, then single-turn GRPO with a verifier-only reward (no live AIDE during RL).

python scripts/batch_run_heuristic.py \
  --csv_path data/ctu_datasets_info.csv \
  --max_tasks 10 --seeds_per_task 2 --steps 20 \
  --data_source hf --upload_hf \
  --out_logs_dir data/heuristic_runs/logs \
  --out_workspace_dir data/heuristic_runs/workspaces \
  --runs_index data/heuristic_runs_index.csv

Upload CTU parquet tables to Hugging Face once: python scripts/upload_ctu_to_hf.py (repo: guilhermedrud/ctu_datasets/data/).

Polyaxon: see polyaxon/AIDE_HEURISTIC.md. Jobs download from HF data/<task>/ and upload logs to runs/<task>/seed<N>/.

python scripts/export_rlhf_data.py \
  --logs_dir data/heuristic_runs/logs \
  --out data/offline_decisions.jsonl \
  --ctu_csv data/ctu_datasets_info.csv

python scripts/prepare_sft_data.py --in_path data/offline_decisions.jsonl --out_path data/sft.jsonl
bash scripts/openrlhf/train_sft.sh

python scripts/prepare_grpo_data.py --in_path data/offline_decisions.jsonl --out_path data/grpo_prompts.jsonl
bash scripts/openrlhf/train_grpo_singleturn.sh

Verifier semantics (aide/rlhf/grpo_verifier.py): reward = logged R_t iff the sampled action matches the logged heuristic action and passes validate_action; 0 if valid but different; -1 if invalid JSON or invalid action.

python scripts/evaluate_policy_offline.py --policy_kind llm --policy_model <your_model_id>

Legacy online multi-turn configs live under configs/openrlhf/legacy/ and scripts/openrlhf/legacy/.

If you use AIDE in your work, please cite the following paper:

@article{aide2025,
      title={AIDE: AI-Driven Exploration in the Space of Code}, 
      author={Zhengyao Jiang and Dominik Schmidt and Dhruv Srikanth and Dixing Xu and Ian Kaplan and Deniss Jacenko and Yuxiang Wu},
      year={2025},
      eprint={2502.13138},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2502.13138}, 
}