synthetic-sciences/openscience

What it solves

OpenScience provides an integrated AI workbench for scientific research, automating the tedious parts of the research loop. It allows researchers to set a high-level goal and have an AI agent handle the literature review, hypothesis formation, coding, experiment execution, and final write-up in a single continuous session.

How it works

The system runs a local server that hosts a browser-based workspace UI, an agent runtime, and a library of scientific tools. A model-agnostic agent plans research tasks using a harness and executes them by calling tools such as a shell, editor, LSP, and specialized scientific connectors. Users can connect various frontier or open-weight models via API keys or local hosting. The agent can delegate tasks to independent "Explore" or "Execute" workers to parallelize work.

Who it’s for

It is designed for researchers and scientists working in machine learning, biology, physics, and chemistry who want an AI collaborator to automate data collection, analysis, and experimentation.

Highlights

• End-to-end research loop: Automates everything from reading papers to writing the final results.
• Scientific database integration: Built-in connectors for UniProt, PDB, Ensembl, ChEMBL, PubChem, arXiv, OpenAlex, and Semantic Scholar.
• Bundled scientific skills: Includes capabilities for training (DeepSpeed, PEFT, TRL), cheminformatics, molecular biology, and LaTeX.
• Integrated workspace: A browser UI featuring a file tree, terminal, editor, and inline rendering for genomes, molecules, and plots.
• Extensible architecture: Supports MCP servers, plugins, a TypeScript SDK, and custom agents.