- Published on
An Open-Source, Model-Agnostic AI Workbench for Machine Learning, Biology, Physics, and Chemistry Research
- Authors
- Name
- DeepHorus
Synthetic Sciences has unveiled OpenScience, an open-source AI workbench designed for research in machine learning, biology, physics, and chemistry. Released under the Apache 2.0 license, it runs on users' own infrastructure, positioning itself as a model-agnostic alternative to Anthropic's Claude Science, which launched in late June 2026. The tool aims to keep scientific AI workflows open and vendor-independent, allowing researchers to swap models and keep data local.
The research team frames OpenScience as a direct response to the growing reliance on proprietary AI tools in science. By releasing it as an independent project, Synthetic Sciences ensures that researchers are not locked into a single ecosystem. The pitch is clear: scientific AI tooling should not be owned by one vendor.
OpenScience provides a browser-based workspace with a local agent runtime that can handle the full research loop. This includes reading relevant papers, forming a hypothesis, writing and running code, conducting experiments, and writing up results. It supports any AI modelâClaude, GPT, Gemini, GLM, Kimi, DeepSeek, or local fine-tunesâon a per-request basis. The tool ships with over 250 editable skills and integrates scientific databases like UniProt, PDB, ChEMBL, and arXiv as agent tools.
Installation is straightforward via npm, and users bring their own API keys. The tool is free and never gated. It features a real workspace with a file tree, editor, terminal, and session history, rendering molecules, structures, genomes, and plots inline. This makes it a practical environment for daily research tasks.
By being open-source and model-agnostic, OpenScience challenges the notion that scientific AI tooling should be owned by a single vendor. It promotes reproducibility and flexibility, as researchers can switch models or run local fine-tunes without changing their workflow. The inclusion of 250+ skills and dozens of database connectors further enhances its utility for interdisciplinary research.
The release is likely to be welcomed by the research community seeking alternatives to proprietary tools. While Synthetic Sciences does not provide direct quotes, the comparison to Claude Science is explicit. The emphasis on openness and user control aligns with the values of many academic and independent researchers.
Looking ahead, OpenScience offers an optional managed layer called Atlas for curated frontier models and cloud compute, but it remains just thatâoptional. With extensibility features like LSP integration, MCP servers, and a TypeScript SDK, the platform could foster a community-driven ecosystem. As scientific AI matures, model-agnostic, open-source tools like OpenScience may become essential infrastructure for reproducible and collaborative research.
