Open Science

Open Science is an open-source desktop application developed by AIPOCH for scientists and researchers who need a reproducible, inspectable AI-assisted research environment. It runs locally on macOS, Windows, and Linux under the Apache 2.0 license, meaning project data, credentials, and artifacts stay on the researcher's computer. The workbench combines agent workflow orchestration, Python and R notebook execution, 24 built-in scientific data connectors, and immutable artifact provenance in a single workspace.

What It Is

Open Science is a local-first AI research workbench — not a chat wrapper or a generic coding assistant. Researchers create a project, describe a task in plain language, and let the agent read files, run code, search the web, call scientific data connectors, and return reports, tables, and figures with traceable provenance. Every action is visible, gated by user approval, and open to review. The application is built on Electron with React, TypeScript, Prisma/SQLite, and an ACP-based agent runtime, and it supports selectable agent backends including Claude Code, OpenCode, Codex, and CodeBuddy.

Core Capabilities

The workbench covers the full research lifecycle from literature review to artifact production:

• Agent workflow — natural-language tasks, streamed responses, tool-activity cards, context compaction, message branching, and subagent delegation
• Execution — persistent Python, R, and REPL kernels with durable code/output history; stateless shell commands; remote SSH compute for HPC clusters; outbound network access limited to approved domains
• Scientific data connectors — 24 built-in connectors including PubMed, bioRxiv, Literature Graph, Genes & Ontologies, Genomes, Variants, Clinical Trials, ChEMBL, ZINC, and more
• Scientific skills — 18 featured file-based research skills including AlphaFold2, ESMFold2, Evo 2, Chai-1, DiffDock, Literature Review, and Remote Compute (SSH); personal and imported skills also supported
• Artifact provenance — immutable, checksummed artifact versions with producer code, execution history, input references, environment inventory, and producing conversation branch
• Preview formats — PDFs, Office documents (DOCX, XLSX, PPTX), images (including TIFF), molecular structures, source code with syntax highlighting, and Notebook history
• Safety controls — three approval profiles (Ask for approval, Auto-approve edits, Full access) with per-connector and per-tool permission policies

Model Agnosticism and Provider Support

Open Science connects to major cloud LLM providers, custom gateways, or existing Claude and Codex subscription logins. Built-in cloud vendors include OpenAI, Anthropic, xAI (Grok), DeepSeek, Zhipu AI (GLM), Kimi (Moonshot), MiniMax, StepFun, SenseNova, Volcengine Ark, Bailian (Alibaba Cloud), Tencent TokenHub, OpenCode, and OpenRouter, among others. Researchers can also supply a compatible Base URL and API key for any custom gateway. Provider and model choices are validated against the selected agent backend, and a per-session composer picker lets researchers choose model and reasoning effort together.

The AIPOCH Ecosystem

Open Science serves as the desktop orchestration layer within the broader AIPOCH ecosystem. It pairs with the AIPOCH medical research skills library — which the vendor describes as containing 550+ reusable skills spanning evidence insights, protocol design, data analysis, and academic writing — and with MedSkillAudit, a domain-specific audit framework that governs whether a medical research agent skill is release-ready before deployment. Skills and connectors remain behind the permission system, and the source code, skill definitions, connector definitions, and artifact provenance are all available for inspection.

Update: v0.24.0 (August 2026)

The latest release, v0.24.0, published August 31, 2026, introduces notebook network sandboxing (limiting notebook and compute access to approved domains with in-conversation approval for new destinations), a project quick switcher with title and description previews, device-wide shared credentials that custom connectors can bind to, immutable file generations for notebook and compute runs, and a complete German interface with matching README translations. The GitHub repository shows the project was created in July 2026 and has been updated actively through September 2026, with the README available in nine languages. The AIPOCH website also reports that Open Science achieved the highest ranking score in the BiomniBench-DA Public 50 compiled comparison, earning a score of 79.05 with gpt-5.6-sol (xhigh).