ACE is a framework for systematic prompt exploration and optimization through atomic concept-level edits. It decomposes prompts into individual concepts and generates targeted mutations (add, remove, replace) to explore how small changes affect model behavior with respect to a given objective. The framework includes:

• ACE Exploration: A tree-search algorithm that iteratively mutates prompts, evaluates them against a scorer that measures the objective (generally including a target model and autorater), and expands the most promising branches.
• Constitution Optimizer: An optimization loop that learns a "constitution" (a set of strategies) to guide the mutation process, using a surrogate classifier to predict which edits will satisfy the objective. This constitution uncovers key aspects of the model's behavior with respect to the objective and can then be used to guide the ACE exploration.

The project uses pyproject.toml for package configuration. To install it in editable mode:

python -m venv ace_env
source ace_env/bin/activate
pip install -e .

To install with all optional dependencies (OpenAI, Anthropic, Hugging Face Datasets, and TensorFlow Datasets):

pip install -e ".[all]"

Simple demo: ace_word_count_demo.ipynb

You can run ACE exploration with a config file:

python run_ace_exploration.py --config=configs/word_count.py

After running an exploration, you can optimize a constitution:

python run_constitution_optimizer.py --exploration_output_dir='<path_to_exploration_output_dir>'

@article{kalibhat2026interpretingcontrollingmodelbehavior,
      title={Interpreting and Controlling Model Behavior via Constitutions for Atomic Concept Edits},
      author={Neha Kalibhat and Zi Wang and Prasoon Bajpai and Drew Proud and Wenjun Zeng and Been Kim and Mani Malek},
      year={2026},
      eprint={2602.00092},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2602.00092},
}

Copyright 2026 Google LLC

All software is licensed under the Apache License, Version 2.0 (Apache 2.0); you may not use this file except in compliance with the Apache 2.0 license. You may obtain a copy of the Apache 2.0 license at: https://www.apache.org/licenses/LICENSE-2.0

All other materials are licensed under the Creative Commons Attribution 4.0 International License (CC-BY). You may obtain a copy of the CC-BY license at: https://creativecommons.org/licenses/by/4.0/legalcode

Unless required by applicable law or agreed to in writing, all software and materials distributed here under the Apache 2.0 or CC-BY licenses are distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the licenses for the specific language governing permissions and limitations under those licenses.

This is not an official Google product.