Open-Science is an open-source, local-first AI research workbench from AIPOCH: a desktop application that takes a scientific project from research question to traceable artifact in one workspace. It combines coordinator and specialist agents, Python and R execution, scientific connectors and reviewer checks, keeping every output connected to the evidence behind it. Released under the Apache-2.0 license, the project has more than 5,300 stars on GitHub and is under active development in 2026.
Scientific work with AI agents usually scatters across chat windows, notebooks and files, and the provenance of a result is lost. Open-Science is built around traceability instead: persistent projects hold the question, files, code, outputs and review evidence together, and agents work through a define–plan–execute–inspect–share loop. Persistent Python and R kernels keep analytical state across steps, shell commands and remote SSH compute are available when local resources run out, and 24 built-in scientific connectors cover literature, biomedical databases, genomics, chemistry and clinical research. Purpose-scoped Specialist agents can be delegated sub-tasks under the researcher’s permissions, and reusable Skills package workflows such as literature review. Every artifact is versioned with provenance records, optional Reviewer checks flag unsupported claims, and selected records export as portable .science packages. The workbench is model-agnostic, connecting to providers from OpenAI and Anthropic to DeepSeek and local gateways, on macOS, Windows and Linux.
Free. Open-Science is open source under the Apache-2.0 license with no seat license; external model or compute providers you connect may charge separately.
Researchers and scientific teams who want AI agents to help with literature, analysis and reporting while keeping a reproducible, auditable record of how every result was produced.