Skip to content
AgentOpen-source agent

chemcrow-public

Open source LLM agent that uses chemistry tools like RDKit and PubChem for chemistry tasks

Pricing
Open source
Platforms
API, self-hosted, Linux
GitHub stars
971
Last commit
2 years ago
Language
Python
License
MIT

Similar agents

About chemcrow-public

chemcrow-public is an open-source agent for ai models and llm tooling. It is available on API, self-hosted and Linux.

The README describes ChemCrow as a Python package for working through reasoning-heavy chemistry questions. Rather than relying on a language model alone, it wires an LLM to a set of chemistry tools, including RDKit for molecular operations, paper-qa for literature lookups, and chemical databases such as PubChem and chem-space.

The package is built on LangChain. A user creates a ChemCrow agent, picks a model such as gpt-4-0613, and sends it a natural-language request; the agent decides which tools to call. Optional SerpAPI access is supported for web search, and retrosynthesis and reaction prediction default to the hosted RXN4Chem API, with Docker images offered for self-hosting those two services on a GPU machine.

The README is explicit that the public repo omits some tools from the original paper because of API usage restrictions, so results will not match the paper. It suits chemists and developers who want a scriptable chemistry assistant they can run and modify themselves.

What it does

Chemistry tool integration
Combines an LLM agent with RDKit, paper-qa and PubChem-style database lookups for chemistry questions.
Retrosynthesis and reaction prediction
Uses the RXN4Chem API by default for retrosynthetic planning and reaction product prediction.
Self-hosted chemistry services
Provides Docker images so the reaction prediction and retrosynthesis services can run on your own GPU.
LangChain-based agent
Built on LangChain, so the agent chooses among tools based on the user's request.
Pip-installable Python package
Installs with pip install chemcrow and is driven from a short Python session.

Good for

  • Answering molecular property questions such as molecular weight of a named compound
  • Predicting the products of a described chemical reaction
  • Planning retrosynthetic routes for a target molecule
  • Running literature-backed chemistry queries through paper-qa

Pricing

Open source

MIT-licensed open source package; uses paid third-party APIs such as OpenAI.

chemcrow-public FAQ

How is ChemCrow installed?

The README gives pip install chemcrow as the installation command.

Does the public package include every tool from the paper?

No. The README says some tools were left out because of API usage restrictions, so results will not match the paper.

Can the reaction prediction run without an external API?

The README says reaction prediction and retrosynthesis can be self-hosted with provided Docker images, with local_rxn=True enabled, and that these need a GPU.

Is chemcrow-public free?

Yes. chemcrow-public is open source, so you can use and self-host it for free.

What are the best alternatives to chemcrow-public?

The closest alternatives to chemcrow-public are llm_agents, DemoGPT, AgentPilot, langchain and openagent.

Where this comes from

  • Description: written from the project's README.