AIPOCH Releases Open-Source AI Research Workbench Open-Science 1.0

By AI123

On October 15, 2026, the AIPOCH team released AIPOCH Open-Science 1.0, an open-source AI research workbench distributed under the Apache-2.0 license. The desktop application supports macOS, Windows, and Linux, with a command-line tool also available. Co-authors from 76 universities and institutions around the world participated in testing and validation, and a preprint has been posted on arXiv.

Open-Science is designed to organize literature, data, code, compute environments, and research outputs within a single project workspace. The execution layer supports Python and R environments, and running kernels can retain variables and analysis state for continued work; long-running tasks can be processed in the background. When local compute is insufficient, researchers can use configured SSH remote hosts or Slurm clusters. The workbench also supports multiple model services and compatible custom gateways, allowing researchers to choose models per session and adjust inference settings.

The platform includes research connectors for scientific literature and biomedicine, as well as curated skills for literature review and computational biology. Researchers can package workflows as skills, install and update skills through a marketplace, and configure expert agents with defined roles and tools. To support review and reproducibility, the workbench records file versions and provenance, linking files to code, inputs, execution history, environment information, and dialogue context. A review feature can surface unsupported conclusions, inconsistencies, or evidence gaps.

Project files and workspace records remain local by default. When cloud models, web search, or remote research connectors are invoked, prompts, necessary context, or task inputs may be sent to corresponding services. Using .science packages, AI-assisted workflows can be transferred between machines as inspectable, handoff-ready archives, enabling collaboration and reuse of established methods.

Explore topicsAIAIPOCHModels

Source

View all