aipoch/open-science

Project overview

By anchoring outputs to immutable provenance records and running on the user's own computer, open-science addresses the context loss that occurs when research is fragmented across disconnected tools.

Project type

AI Agent · RAG · Data Processing

Use cases

Knowledge Q&A · Search & Research · Data Analysis

Deployment

Refer to project documentation

License

Apache-2.0

Best for

• Researchers who need an inspectable desktop workspace that keeps results connected to the code, data, and execution history that produced them.
• Teams that want local-first project state with the flexibility to choose their own cloud model provider, custom gateway, or existing Claude or Codex subscription.
• Developers and data teams who want to drive research tasks via a headless CLI or Node.js SDK that shares the same daemon and sessions as the desktop interface.

Key capabilities

• Create, manage, and persist research projects and sessions. Drafts, files, previews, and run history survive application restarts. Completed prompts can be edited into persistent, selectable message branches without deleting the original downstream path.
• The AI agent can run commands, Python and R code, edit files, search the web, call scientific data connectors, and generate artifacts from natural-language tasks. Execution is presented with typed tool-activity cards, approval pauses, and stop controls.
• Generated artifacts are stored as immutable, checksummed versions. A provenance view exposes available producer code, execution history, input references, environment inventory, conversation branch context, and reviewer evidence, explicitly marking unavailable evidence.
• Includes a catalog of file-based research skills that can be created, imported, and used. Built-in life-science connectors for literature, genes, proteins, and other domains are provided, alongside support for custom local and remote MCP connectors exposed as permissioned agent tools.
• Provides responsive multi-tab previews for common scientific data, Office documents, images, source code, molecular structures, and Notebook history directly in the workspace without truncating the underlying file.
• An opt-in reviewer audits a completed turn against its transcript, execution log, and artifacts to report pass, warn, or fail findings and run a bounded fix loop to correct issues.
• Connects to major cloud LLM providers, custom compatible gateways, or reuses existing Claude or Codex subscriptions. Model and reasoning effort selection is available in the composer.
• Provides a headless CLI and a zero-dependency Node.js SDK that use the same local daemon, projects, sessions, credentials, and permissions as the desktop interface.

Limitations and risks

• Deterministic reconstruction, portable environment restoration, and full-fidelity session replay are not yet available.
• Generated output does not replace expert judgment, statistical review, or validation against primary evidence.
• External data flow is possible: model requests, web searches, and remote connectors may send prompt content, context, and parameters to external services. These calls happen only through services the user explicitly configures or approves.

Getting started

• Download the installer for macOS, Windows, or Linux from the latest release page, run it, and launch the app. The app-managed runtime can be installed without requiring Node.js or an administrator password. Complete the guided Prepare environment step and the Model provider configuration to finish setup.

Evidence and sources

• GitHub project description: Open Science is an open-source, local-first, model-agnostic AI research workbench for scientific discovery.
• README: Open Science is available as a released desktop application and is actively developed.
• README: Research work is usually split across chat windows, notebooks, local scripts, scientific databases, file browsers, and reporting tools. Context is lost at every handoff, and the a…
• README: Explore alternative research paths by branching from an earlier message without losing the original conversation, and inspect every generated result as an immutable version with p…
• README: Open Science is an independent product built from scratch. It is not a proxy, unofficial client, or reskin of another AI research application.

Source updated at: 2026-07-30T01:10:32Z

Open original GitHub project