Open-Science
Organize projects, run analyses, and inspect outputs.
Orchestrate end-to-end scientific workflows with agents, tools, and data connectors in one place.
System roleWorkbench · Orchestrator
AIPOCH PRODUCT/ OPEN-SOURCE RESEARCH WORKBENCH
Science, Open to All
Open-Science is AIPOCH's open-source, local-first AI research workbench. It combines agent workflows, Python and R execution, scientific data connectors, and traceable research artifacts in one inspectable workspace.
25OFFICIAL MODEL APIs
+ CUSTOM GATEWAY
+ CUSTOM GATEWAY
4AGENT FRAMEWORKS
548marketplace
+ built in
+ built in
36SCIENCE CONNECTORS
GPT
Claude
Grok
DeepSeek
Qwen
GLM
Kimi
MiniMax
StepFunProduct tour / Video
See Open-Science in Action
Find installation steps, project setup, workspace guides, and troubleshooting. Check the documented version when following instructions.
Ecosystem / signal flow
The AIPOCH ecosystem
Scientific AI workflows
Open Science runs and documents the work. Medical Research Skills provide vetted domain methods, and MedSkillAudit checks each skill before release.
Medical Research Skills
Domain knowledge and execution logic.
Access 550+ vetted medical research skills covering analysis, reporting, data ops, and study operations.
Skill library597 skills
1.9K
MedSkillAudit
Audited release-ready before deployment.
Validate skills for safety, correctness, and reliability with automated and human-in-the-loop review.
Release gatePass / Reject
See What’s New
in Open-Science
Find installation steps, project setup, workspace guides, and troubleshooting. Check the documented version when following instructions.
Open the Wiki
Latest release
Oct, 2026Open-Science v1.0.0
A local-first AI research workbench unifying literature, scientific data, agent-assisted analysis, inspectable results, and portable .science research packages across macOS, Windows, and Linux.
- Portable research packages
Export selected conversation branches, file versions, Notebook records, and verification evidence as .science packages for review, handoff, and archiving.
- Integrated literature workspace
Organize references, screen papers, annotate PDFs, extract figures and tables, and export bibliographies from one workspace.
- Flexible research agents
Use supported agent frameworks, model providers, research skills, specialists, and scientific connectors with scoped permissions.
- Python, R, and remote compute
Run persistent Notebook environments locally or connect to configured SSH hosts and Slurm clusters.
- Inspectable results
Preserve artifact versions with code, inputs, execution history, environment evidence, optional review, and eligible re-execution.
Articles & guides
Read & watch
AIPOCH Open-Science Leads the Reported ASI-Bench B1–B4 ComparisonsProduct&Architecture · Oct 15, 2026AIPOCH Launches Open-Science 1.0: Every Discovery, a New BeginningProduct&Architecture · Oct 14, 2026AIPOCH Open-Science v1.0.0: What Ships and How to UpgradeRelease notes · Oct 13, 2026AIPOCH Open-Science v0.37.2: Runtime Recovery, Linux CLIRelease notes · Oct 12, 2026AIPOCH Open-Science v0.37.1: Faithful PDFs, Stronger WindowsRelease notes · Oct 11, 2026
All posts Product overview
What it does
- Package and continue with .science — Carry selected conversation branches, file and artifact versions, Notebook records, and verification evidence to another project or computer for review, handoff, or archiving. Import packages as read-only records, or create a writable copy to continue the work.
- Open architecture — Connect multiple model providers, agent frameworks, custom gateways, reusable Skills, and scientific connectors in an Apache-2.0 codebase.
- Execute research workflows — Run Python, R, and shell workflows directly so agents can perform computational tasks, not only suggest code.
- Manage the literature — Import references by identifier or file, organize collections, compare and merge duplicates, retrieve open-access PDFs where available, and format citations.
- Preserve traceable artifacts — Keep reports, tables, figures, and notebooks as immutable versions with available links to inputs, code, execution history, environment details, conversation branches, and reviewer evidence.
- Control the infrastructure — Keep project state on your machine by default and govern file, command, network, connector, and compute access through permissions. Connect locally deployed models or external services when configured.
The Workbench / Workflow
Run and Review Research
in Open-Science
Create a project, add your research files, and describe the task. Use AI agents to run analysis and create reports, tables, and figures, then inspect the outputs alongside available execution records and provenance.
Describe the task in plain language. The agent drafts a step-by-step execution plan for your review — nothing runs until you approve it.
Learn moreRun commands, Python, R, search, and connectors in one persistent workspace. Every action stays visible, approval-gated, and open to review.
Learn moreTurn the work into reports, tables, figures, and structured outputs while preserving the evidence and steps behind every result.
Learn moreOpen every result in place. Inspect data, documents, images, source, structures, and notebook output before you approve or share it.
Learn more
Agent Skills / The library
597 skills
Medical Research Agent Skills
Browse AIPOCH skills for literature review, study planning, data analysis, and academic writing. Each skill packages instructions and supporting resources for a specific task. Check its requirements and any available evaluation results before use.
01
Evidence Insights
Search, assess, and synthesize research evidence.
02
Protocol Design
Plan study methods, endpoints, and analyses.
03
Data Analysis
Prepare data, run analyses, and inspect results.
04
Academic Writing
Draft manuscripts, methods, and figure descriptions for researcher review.
Claude Code
Codex
OpenCodeBenchmark / Release gate
MedSkillAudit
Evaluate Medical Research Agent Skills
MedSkillAudit is a domain-specific audit framework to evaluate whether a medical research agent skill is ready for release before deployment. It combines two hard veto gates with a two-layer scoring system: static design evaluation and dynamic task-based testing.
Scientific Integrity
No fabricated citations, DOIs, sample sizes, or p-values. Enforced as a hard veto gate — one FAIL means rejection, regardless of the numeric score. Applies to research skills in categories 1–4.
Reproducibility
Core results must stay reproducible: no unseeded random-number calls, no unbounded loops, and generated code that runs with its declared dependencies. Each audit runs 3, 5 or 7 test inputs, scaled to assessed complexity.
Methodological Validity
No logical fallacies such as conflating correlation with causation; statistical methods and study-design choices must be valid and defensible for the stated task. Also a hard veto gate — one FAIL means rejection, regardless of score.
