AIPOCH launches Open-Science 1.0, a local-first research environment with scientific agents

AIPOCH has released version 1.0 of Open-Science, an open-source, local-first, model-agnostic workbench for AI research. The tool targets researchers and scientists who want to run complete research workflows — literature review, hypothesis development, code execution, data analysis, and the production of reports, tables, and figures with full provenance tracking — without sending sensitive data to the cloud. It runs on macOS, Windows, and Linux, installable from the official downloads page or via Homebrew on Mac.

At its core are scientific agents that can read files, search the web, execute Python or R code in a shared notebook, query scientific data sources, and generate outputs with audit trails. The researcher defines a goal in natural language, attaches files or references existing artifacts using @, selects a model and an approval mode, and the agent combines skills, authorized connectors, searches, and code execution. Every tool action is logged and viewable in a Provenance window, so any number or figure can be traced back to its origin.

A central feature in 1.0 is the ability to export sessions as portable researchers in the portable.science format — packages that contain selected conversation branches, file versions, notebook records, and verification evidence. The aim is to enable peer review, handoff between researchers, and long-term archiving without dependence on a specific platform. The current release unifies the local workbench, the agents, Python and R notebook execution, literature tools, and the portable packages, a jump from the preceding 0.37.2.

A setup wizard walks through five steps: environment, data location, agent runtime, model provider, and notebook runtime. An environment check and connection to the chosen model are required; notebook settings and data location can be changed later. On Windows, reinstalling preserves research data, and a reset tool wipes the local installation after confirmation. The standalone CLI version bundles Node and runs without the desktop app, with shared profiles and updates managed through a separate AIPOCH tap in Homebrew.

A representative scenario demonstrates the flow: reproducing a published differential-expression analysis in bioinformatics, comparing the reproduced results to the original, and generating the report, tables, and figures needed for review. The researcher describes the research question, the paper, the datasets, required methods and thresholds, expected deliverables, and acceptance criteria. The agent weaves together scientific skills, authorized connectors, searches, file operations, and notebook code, with every step documented and evidence available for inspection.