GitHub
FreeOpen source package for accurate solution of reasoning-intensive chemical tasks
About GitHub
ChemCrow is an open source package for the accurate solution of reasoning-intensive chemical tasks. Built with Langchain, it uses a collection of chemical tools including RDKit for chemical informatics, paper-qa for literature search, and integrates with databases like Pubchem and chem-space. The package is based on the research paper 'ChemCrow: Augmenting large-language models with chemistry tools' (arXiv:2304.05376). Note that this repository does not contain all tools described in the paper due to API usage restrictions, and results may differ from the paper. ChemCrow supports self-hosting for retrosynthetic planning and reaction product prediction via Docker. It is licensed under MIT and available on HuggingFace.
Key Features
Pros & Cons
- Accurate solution of reasoning-intensive chemical tasks
- Modular and extensible via Langchain
- Integrates multiple chemical databases and tools in one package
- Open source with permissive MIT license
- Supports self-hosting for critical tools to reduce external API dependency
- Free to use (requires own OpenAI API key)
- Does not include all tools from the ChemCrow paper due to API restrictions, limiting reproducibility
- Requires valid OpenAI API key for core functionality
- Default retrosynthesis and reaction prediction rely on external RXN4Chem API (can be slow)
- Self-hosting requires Docker and GPU, adding setup complexity