openscience: an AI workbench for scientific research that automates the full loop from literature review to experiment execution

What it solves

OpenScience is an AI-powered workbench designed to automate the scientific research loop. It removes the manual effort of switching between reading literature, forming hypotheses, writing code, running experiments, and documenting results by integrating these steps into a single, continuous session.

How it works

The system runs a local server that manages a browser-based workspace UI, an agent runtime, and a tool layer. Users provide their own API keys for frontier or open-weight models. The AI agents plan their research using a research harness and execute tasks via a set of tools, including a shell, editor, LSP, and specialized scientific connectors. All work, including session history and artifacts, is stored locally on disk.

Who it’s for

It is built for researchers in fields such as machine learning, biology, physics, and chemistry who want an AI collaborator that can interact with real compute and scientific databases.

Highlights

• End-to-end research loop: Automates literature review, hypothesis generation, code execution, and final write-ups.
• Specialized agents: Includes a general research agent and specialists for biology, physics, and ML, supported by critique and literature-review sub-agents.
• Extensive toolset: Features over 290 skills across training, evaluation, and cheminformatics, plus direct query access to 30+ scientific databases like UniProt, PDB, and arXiv.
• Integrated workspace: A browser UI providing a file tree, terminal, editor, and inline rendering for genomes, molecules, and plots.
• Model-agnostic: Compatible with providers like Anthropic, OpenAI, and Google, as well as local models.