OpenScience is an open-source AI workbench for scientific research that automates the execution of research tasks while keeping every step visible and under human control.
The tool addresses the problem of automating routine but essential research work—literature review, data analysis, experiment reproduction, and report drafting—while preserving human judgment over the research direction and conclusions. It works by accepting a plain-language description of a research goal, then planning the approach, gathering evidence from scientific sources, writing and running code and experiments, and presenting results with a complete audit trail of what it did at each step. The agent can delegate bounded subtasks to worker agents running in parallel while maintaining synthesis and final decision-making at the lead level.
Researchers conducting computational work in biology, chemistry, physics, machine learning, or data engineering should consider this tool, particularly those spending significant time on data cleaning, parameter sweeps, literature searches, and methods documentation. The workbench runs as a desktop application, browser interface, or terminal command on your own machine against your files, with access to real execution environments including shell, Python, and R kernels, plus hundreds of bundled scientific skills and connectors to databases like ChEMBL, UniProt, PubMed, and arXiv. You can bring your own API keys, use a supported provider, run a local model, or use the managed pay-as-you-go option called Ace.
The project shows active development with regular updates to the desktop application and ongoing expansion of scientific capabilities. The codebase is written in TypeScript and includes comprehensive documentation covering installation across macOS, Windows, and Linux platforms, along with quickstart guides and contribution guidelines. The tool emphasizes transparency through its design philosophy that every operation must be reviewable after execution, reflecting a commitment to maintaining scientific rigor and auditability in automated research workflows.
