OpenScience - Synthetic Sciences: An open-source AI research workbench

What is OpenScience?

OpenScience is an open-source AI research workbench launched by Synthetic Sciences, targeting research in machine learning, biology, physics, and chemistry. The platform supports the complete research cycle from literature review, hypothesis generation, code writing, experimental execution to results analysis and paper writing. It supports switching between arbitrarily large models on request and includes over 250 editable skills and approximately 30 scientific databases as intelligent agent tools. Users can run it for free on their local infrastructure with their own API keys, and it can be installed with a single click via npm without account registration.

Main functions of OpenScience

• Complete scientific research cycleIt covers literature retrieval, hypothesis generation, code writing, experiment execution, data analysis, and paper writing.
• Model-independent routingSupports Claude, GPT, Gemini, GLM, Kimi, DeepSeek, and local fine-tuning models, allowing for seamless switching upon request.
• 250+ editable skillsIt covers training (DeepSpeed, PEFT, TRL), evaluation, dataset processing, cheminformatics, LaTeX, graphing, and cloud computing.
• Scientific database toolsIt integrates approximately 30 databases, including UniProt, PDB, Ensembl, ChEMBL, PubChem, arXiv, OpenAlex, and Semantic Scholar.
• Professional intelligent agent: Research, biology, physics, machine learning agents, and critical review and literature review sub-agents.
• Built-in workspaceFeatures include: file tree, editor, terminal, session history, and support for inline rendering of molecules, structures, genomes, and graphs.
• Scalable architectureSupports LSP integration, MCP server, plugins, and TypeScript SDK.

Follow us on WeChat and reply with "open source",join inAI open source project discussion group

How to use OpenScience

• InstallInstall npm packages globally npm install -g @synsci/openscience.
• start up:run openscience The command opens the workspace in the browser.
• Select ModelFor the first run, choose the Atlas managed model, the one with the vendor key, or the free demo model.
• Set goalsEnter your research goals in the workspace, and the agent will automatically advance the process of literature review, hypothesis development, code creation, experimentation, and writing.
• Switching modelsIn the model selector, you can switch between any provider or local model on request without changing the configuration.
• Extended customizationCustomize skills and agents using LSP, MCP servers, plugins, or the TypeScript SDK.

OpenScience's core advantages

• Fully open source and auditableIt adopts the Apache 2.0 license, all code and skills are open and readable, the scientific research reasoning chain is transparent and reproducible, and the risk of black box is completely avoided.
• Model-independent zero-lockingIt supports seamless switching between Claude, GPT, Gemini, DeepSeek, and local models upon request, and new models can be seamlessly integrated upon release, without being tied to a single vendor.
• Data localization complianceIt runs on its own infrastructure, keeping all private datasets and API keys locally, meeting the compliance requirements of keeping sensitive data within the domain, and comes with free keys with no access restrictions.
• Tools covering industry leadersIt features over 250 editable skills and approximately 30 scientific databases, offering a breadth of tools far exceeding similar closed-source products and supporting continuous expansion.
• Deep integration of professional work areasIt includes a complete browser workspace with a file tree, editor, and terminal, supports inline rendering of molecular structures and genome maps, and seamlessly integrates research workflow with code environment.

OpenScience project address

• Project official websitehttps://www.openscience.sh/
• GitHub repositoryhttps://github.com/synthetic-sciences/openscience

Comparison of OpenScience with similar products

Application scenarios of OpenScience

• Machine learning researchThe ML agent pulls arXiv papers, uses PEFT and TRL skills to write training scripts, perform fine-tuning, and generate reports.
• Computational biologyThe biological agent queries UniProt and PDB, renders the protein structure inline, proposes candidate mutation schemes, and records their sources.
• CheminformaticsThe agent queries ChEMBL and PubChem to obtain bioactivity data, runs a screening filter, and returns ranked candidate molecules and charts.
• Model Comparison Under Budget ConstraintsThe team runs the same task on Claude, GLM, and local models, switching between them with a single click to compare cost and quality.