Synthetic Sciences Launches OpenScience AI Workbench

Synthetic Sciences has launched OpenScience, an open-source AI research workbench under the Apache 2.0 license. The tool runs locally via a browser interface and is designed to automate the entire scientific research loop. Users can instruct an AI agent to search literature, formulate hypotheses, write and execute code, query databases, run experiments, and compile results. The software can be installed using the commands "npm install -g @synsci/openscience" or "npx synsci".

The workbench is model-agnostic, allowing researchers to use their own API keys for models from Anthropic, OpenAI, Google, and various open-weight alternatives. OpenScience ships with more than 250 specialized skills, including DeepSpeed, PEFT, TRL, LaTeX, and cheminformatics tools. It also features direct connectors to roughly 30 databases, such as UniProt, PDB, Ensembl, ChEMBL, PubChem, arXiv, OpenAlex, and Semantic Scholar.

To handle diverse tasks, the platform includes a general research agent alongside specialized agents for biology, physics, and machine learning. It also features critique and literature-review sub-agents, as well as a read-only planning mode. All sessions and generated artifacts are stored locally on the user's machine. While the bring-your-own-key model is entirely free, Synthetic Sciences also offers a managed Atlas platform that provides prepaid access to frontier models and cloud computing resources.

For AI practitioners and scientific researchers, OpenScience consolidates tools that previously required separate coding agents, database clients, and model gateways. By integrating these disparate systems into a single browser-based workspace, the platform streamlines the workflow for complex scientific tasks. However, developers note that human researchers must still verify the generated methods, citations, code, and final conclusions to ensure accuracy.