Open Science Desktop

Open Science Desktop is an AI research workbench for researchers who want to keep their project files locally and inspect how an agent reached its results. It runs on macOS, Windows and Linux, or on a self-hosted server with browser access. Its research agents connect literature searches, experiment code and analysis to a finished manuscript.

The outputs are inspectable files. Figures, notebooks and reports link back to the code, inputs, environment and conversation that produced them. Run records cover local execution, SSH/Slurm jobs, Modal and notebook batches, so remote experiments retain the same research trail.

Model choice comes through OpenCode, including local options, cloud providers and custom OpenAI-compatible endpoints. Sessions, data, notebooks and run records stay in local folders by default; cloud model providers and remote services handle the work you send to them. The authenticated browser gateway keeps API keys on the host machine.

Python and R notebooks run with local kernels. Research connectors include arXiv, PubMed, Crossref and Semantic Scholar, while review skills check traceability, statistical integrity and publication figures. The app can also control Chrome or native research software through accessibility APIs.

Projects retain memory across sessions, and split panes let researchers work with different models side by side. Agent Client Protocol support connects Codex, Gemini CLI and Claude Code to the workbench, or lets Zed, JetBrains and Neovim drive it. Built-in viewers open scientific formats including molecules, 3D meshes, genome tracks and FITS files.