open-science

Open Science Desktop turns the AI-for-science workflow into one desktop app: literature survey, hypothesis, experiment code, figures, and the final write-up in a single auditable session. It's a free, MIT-licensed, local-first workbench for macOS, Windows, and Linux, positioned as the open alternative to hosted AI research products.

You bring your own model keys; the app is model-agnostic and talks through a bundled OpenCode runtime. Everything an agent produces (figures, notebooks, run records) traces back to the exact code and conversation that made it, and runs can target your machine, SSH or Slurm clusters, or Modal. A token-gated gateway can serve the full UI to a browser or your phone.

Researchers and grad students doing computational work are the obvious fit. Sessions, data, and provenance stay in local folders, which matters when your data can't leave the building.

The catch: it's early, version 0.3 early. An agent that drives your real Chrome profile and submits jobs to your cluster is powerful, and exactly the kind of power you should think twice about handing to a young codebase.