Synthetic Sciences Launches Open-Source AI Workbench for Science

Synthetic Sciences launched OpenScience this week, an open-source AI workbench designed for scientific research across machine learning, biology, physics, and chemistry. Licensed under Apache 2.0, the tool is designed to run on users' own infrastructure, positioning itself as an open alternative to proprietary platforms like Anthropic's Claude Science, which was released in late June 2026. OpenScience aims to keep research workflows open, models swappable, and data local, operating as an independent project not affiliated with Anthropic.

The workbench facilitates the entire research process, from literature review and hypothesis generation to code development, experimentation, analysis, and writing. It is model-agnostic, meaning it can integrate with any frontier or open-weight AI model, including Claude, GPT, Gemini, GLM, Kimi, and DeepSeek, allowing users to switch models on a per-request basis. OpenScience includes over 250 editable skills and integrates with numerous scientific databases such as UniProt, PDB, ChEMBL, and arXiv, functioning as agent tools. Usage of the workbench with personal API keys on the user's infrastructure is free and unrestricted.

OpenScience operates as a browser-based workspace supported by a local agent runtime. Users define a research goal, and the AI then follows a collaborative research process. This includes reading relevant papers, formulating hypotheses, writing and executing code, and running experiments. It also queries major scientific databases and compiles results, all within a single, continuous session. The model-agnostic design allows users to leverage their own API keys without requiring an account to begin. Installation is managed via npm with the command `openscience`, which then opens the workspace in a web browser. Alternatively, users can run the tool using `npx synsci` for a single-step execution.