ustc-ai4science/Science-Star
An open-source platform for building, extending, and experimenting with scientific AI agents using a multi-agent architecture.
Not currently ranked — collecting fresh signals.
star history
Science-Star provides a framework for scientific AI agents built on the ReAct engine with modules for Planning, Action, Memory, and Reflection. It includes benchmarks like Humanity’s Last Exam (HLE) and GAIA, and supports a multi-agent architecture combining CodeAgent with search agents. The platform offers tooling for search, web crawling, PDF parsing, and RAG retrieval, enabling researchers and developers to run experiments and extend the system.
Frequently asked
- What is ustc-ai4science/Science-Star?
- An open-source platform for building, extending, and experimenting with scientific AI agents using a multi-agent architecture.
- Is Science-Star open source?
- Yes — ustc-ai4science/Science-Star is open source, released under the MIT license.
- What language is Science-Star written in?
- ustc-ai4science/Science-Star is primarily written in Python.
- How popular is Science-Star?
- ustc-ai4science/Science-Star has 751 stars on GitHub.
- Where can I find Science-Star?
- ustc-ai4science/Science-Star is on GitHub at https://github.com/ustc-ai4science/Science-Star.