Reproducible science needs more than another chatbot

What it does

Open Science is a cross-platform desktop app that turns a plain-language research task into a structured, persistent project. You describe a goal, attach files, and an agent reads data, runs Python or R, searches the web, calls life-science connectors, and generates reports, tables, and figures. All sessions, file edits, tool calls, and previews live in one local workspace that survives restarts.

The interesting bit

The app treats the conversation as a durable lab notebook rather than a disposable chat thread. Every command, file read, and connector call is logged; generated artifacts stay attached to the session instead of drifting in a browser tab. Because it is model-agnostic, you can bring your own provider or gateway rather than renting a locked-in platform.

Key highlights

• Local-first: project state and files stay on your machine; external calls require explicit approval.
• Model-agnostic: supports built-in cloud providers, Anthropic-compatible gateways, or direct Claude subscription tokens.
• Scientific connectors: built-in tools for literature, genomics, proteins, structures, clinical data, chemistry, and drug regulation with per-tool permission controls.
• Extensible skills: add research skills via SKILL.md, ZIP packages, or GitHub imports, and invoke them with / in the composer.
• Native previews: renders CSVs, Office documents, molecular structures, notebooks, and images side-by-side with the conversation history.

Caveats

• macOS and Windows installers currently trigger unidentified-developer or unknown-publisher warnings, so you’ll need to verify the package manually before opening.
• Interface labels and connector catalogs shown in README screenshots may differ from the version you install.

Verdict

Researchers who want a local, inspectable alternative to cloud AI notebooks—with control over models and data boundaries—should look here; if you just need a quick web chat with GPT-4, this is overkill.