Statement on Code and Agent Technology

Acknowledgment

The development of the PARAMUS platform was profoundly inspired by early pioneering work in chemistry agent systems. I extend my sincere gratitude to Andrew D. White, Philippe Schwaller, and their collaborators for releasing the first openly accessible agent codebases—particularly within the ChemCrow project. These foundational efforts offered critical conceptual and architectural insights that significantly influenced the initial design and direction of PARAMUS.

While PARAMUS no longer integrates external agent system code due to incompatibilities with modern design patterns or diverging paradigms, we continue to acknowledge and respect these seminal contributions.

Third-Party & Open-Source Libraries Used

PARAMUS employs two curated chemical safety datasets to support compound risk assessment:

• Clintox dataset from DeepChem, for classification by clinical toxicity and FDA approval status.
• OPCW/Australia Group–based dataset, originally curated by ChemCrow, for safety screening.

In the program itself, under the „Licenses“ menu item, all libraries, their authors, and the respective license models are listed. As this list changes frequently, please refer to the program directly for the most up-to-date information.

Legacy Code

In legacy version 1.x of the PARAMUS Open Cheminformatics Agent (until 03/2025), components of the following agent systems were included:

• ChemCrow, from Bran et al., 2023 [arXiv:2304.05376]. MIT License.
• CACTUS (Chemistry Agent Connecting Tool-Usage to Science), described in McNaughton et al., 2024 [arXiv:2405.00972]. BSD 2-Clause „Simplified“ License.
• ChemAgent: Tooling or Not Tooling? The Impact of Tools on Language Agents for Chemistry Problem Solving, Yu et al., 2024 [arXiv:2411.07228]. MIT License.
• MDCrow: Automating Molecular Dynamics Workflows with Large Language Models, Campbell, Q., Cox, S., Medina, J., Watterson, B., & White, A. D. (2025). arXiv:2502.09565 [cs.AI]. MIT License.

We do utilize widely recognized open-source libraries – such as Dash – to power specific functionalities, uphold best practices, and enhance the overall reliability of our platform.