ai4s-research/open-science

What it solves

Open Science Desktop is a local-first, model-agnostic AI research workbench designed to automate the end-to-end scientific research loop. It replaces fragmented workflows by integrating literature surveys, hypothesis generation, experiment coding, data analysis, and paper writing into a single, auditable desktop environment where every result can be traced back to its origin.

How it works

The platform uses a bundled ai4s-agent that chains specialized skills (like research-explorer and paper-writer) to produce tangible artifacts—such as PDFs, CSVs, and notebooks—rather than just chat responses. It is built with Tauri and React, utilizing a model-agnostic SDK and a pinned OpenCode sidecar to allow users to bring their own LLMs. The system maintains a strict provenance record (provenance.jsonl) to link every figure or report to the exact code and model output that created it.

Who it’s for

It is designed for scientific researchers who need an autonomous agent to handle the heavy lifting of literature reviews and experimental analysis while maintaining full auditability and reproducibility of their results.

Highlights

• End-to-End Research Loop: Automates the process from broad exploration to a finished, publication-grade paper.
• Local-First Privacy: Data, sessions, and provenance records are stored locally by default.
• Full Provenance: Every artifact (figure, table, report) is linked to the specific code, inputs, and conversation that produced it.
• Headless Mode: Includes a CLI (osd) and server mode for running the workbench on remote research machines without a screen.
• Browser Control: Agents can drive the user's own Chrome browser (preserving logins) to harvest live web data.
• Extensible Ecosystem: Supports MCP (Model Context Protocol) connectors for scientific databases like PubMed, arXiv, and the Materials Project.
• Multi-Pane Interface: Supports split-pane tiling with independent models and reasoning efforts per pane.