openscience

Why we included this project

Hand OpenScience a research goal and it carries it through the full research loop in a browser session: reading the relevant papers, forming a hypothesis, writing and running code, querying scientific databases like UniProt, PDB, Ensembl, and arXiv, and closing with a written summary. You connect your own model API keys, and since it accepts any frontier or open-weight model, nothing ties you to a single vendor. Beyond the default research agent there are specialists tuned for biology, physics, and machine learning. The 290+ built-in skills cover training stacks like DeepSpeed and PEFT, dataset preparation, and cheminformatics, among other research tasks, so the tool runs real experiments rather than acting as a chat wrapper. For researchers and small teams who want a reproducible assistant on their own machines with their own keys, this handles the repetitive parts of the research loop in one place.