Curie is an AI-agent framework designed for automated and rigorous scientific experimentation, targeting researchers and scientists. It aims to accelerate the research lifecycle by automating tasks from hypothesis formulation to result interpretation, enhancing precision, reliability, and reproducibility.

How It Works

Curie employs AI agents to manage the end-to-end experimentation process. It breaks down complex research questions into manageable steps, including hypothesis generation, experiment design, execution, and analysis. The framework emphasizes rigor through built-in verification modules that ensure methodical procedures and reproducible results.

Quick Start & Requirements

• Installation: pip install curie-ai
• Prerequisites: Docker, API keys (e.g., Anthropic API key). GPU is recommended for ML tasks.
• Setup: Requires configuring API keys via curie/setup/env.sh.example.
• Resources: Experiments can take several minutes; use of tmux is recommended.
• Documentation: Tutorial, Demo Video

Highlighted Details

• Automated experimentation from hypothesis to reflection.
• Rigor enhancement with built-in verification modules.
• Broad applicability across ML engineering, system analysis, and scientific discovery.
• Supports user-provided starter code and datasets.

Maintenance & Community

• Active development with recent publications on EXP-Bench and AutoML features.
• Community support via GitHub Issues and a Slack Community.

Licensing & Compatibility

• Licensed under the Apache 2.0 License.
• Permissive license suitable for commercial use and integration with closed-source projects.

Limitations & Caveats

• The framework is designed for terminal use; Jupyter Notebooks are not supported.
• Experiment duration can be significant, requiring careful resource management.