Open Science is a local desktop AI workbench designed for researchers, facilitating model-agnostic scientific discovery. It addresses the fragmentation of research tools by consolidating tasks, execution, and artifact generation into a single, inspectable workspace. The application empowers researchers to describe goals in natural language, enabling an agent to read files, execute code, search the web, and leverage scientific data connectors, ultimately producing reports, tables, and figures with a traceable activity history.

How It Works

The core approach is a "plan → execute → produce → preview" workflow managed within a local desktop application. Researchers define projects and sessions, describing tasks in plain language. An agent then interprets these tasks, interacting with various tools including file system access, code execution (Python Notebook kernel), web search, and specialized scientific data connectors. Key design principles include being "open by default" with inspectable code and formats, a "local-first" data strategy ensuring project state resides on the user's machine, and "human-in-the-loop" controls for sensitive actions. This design prioritizes user autonomy, data privacy, and research reproducibility.

Quick Start & Requirements

Installation involves downloading platform-specific installers (macOS DMG, Windows x64 installer, Linux AppImage/Debian) from the latest release page. First-time setup includes guided steps for environment compatibility checks and model provider configuration (supporting built-in cloud providers, Anthropic-compatible custom gateways, or Local Claude). Prerequisites for development include Node.js LTS or newer with npm and Git. Python 3 is optional but required for the built-in Notebook kernel. Setup time is minimal, focusing on environment and model connection validation.

Highlighted Details

• Unified workspace for projects, sessions, files, artifacts, and conversation history.
• Extensible through user-created or imported "skills" (SKILL.md/.ZIP/.skill packages) and custom connectors.
• Comprehensive catalog of life-science connectors for literature, genomics, clinical research, and more.
• Local-first data management ensures project state, settings, and credentials are stored locally.
• Granular safety controls (Ask for approval, Auto-approve edits, Full access) govern agent actions.

Maintenance & Community

The project is actively developed, with installers published on the GitHub releases page. Community engagement is facilitated through GitHub Issues for bugs/features, GitHub Discussions for design conversations, and a Discord server for coordination and help. Release announcements and updates are shared via X (@aipoch_ai).

Licensing & Compatibility

Open Science is distributed under the Apache License 2.0. This permissive license allows for commercial use and integration into closed-source projects, provided attribution and license terms are met.

Limitations & Caveats

While the agent executes tasks, researchers remain ultimately responsible for scientific validity, interpretation, and data privacy. External calls to model providers or web services transmit data, requiring user review of sensitive inputs and provider policies. The application is not a replacement for expert judgment or rigorous statistical validation.