OrganUI

OrganUI (organui.com) is a component registry in the style of shadcn/ui: you install source files into your project with the shadcn CLI and own them afterwards. Its subject is medicine and biology: anatomy, imaging, molecules, clinical widgets. Checked 2026-10-22 against the README, docs/architecture.md, AGENTS.md, SECURITY.md, THIRD_PARTY_NOTICES.md, registry.json, the packages/ layout and the GitHub API. The default branch is master. I did not install any item, open the site, or run the viewers.

What is in the registry

Install, from the README:

bunx shadcn@4.22.1 add @organui/heart
bunx shadcn@4.22.1 add https://organui.com/r/molecule-viewer.json
bunx shadcn@4.22.1 add organui/organui/anatomy-tree     # straight from the GitHub repo

@organui is in shadcn's registry directory, so no components.json entry is needed for the first form. Icons are IconPlaceholder calls, which the CLI swaps for the project's own icon library, so no icon package is imposed.

Design points worth knowing

• Source distribution, not an npm runtime. architecture.md says future versioned runtime packages should only appear if a shared, stable contract is justified. A fix reaches you only when you re-run shadcn add for that item (SECURITY.md).
• Organ models are not in the repo. Most organ src URLs point to jsDelivr, pinned to a full commit of a separate "explorer" repository. The sources named are BodyParts3D (CC BY-SA 2.1 JP or CC BY 4.0), Z-Anatomy (CC BY-SA 4.0), HuBMAP (CC BY 4.0) and a CC0 skin model. Organs with no source there, starting with nerves, are built in-repo and served as GLB from organui.com. The GLB loads at runtime from those hosts and keeps its own license, and the viewer shows the credit by default. If you ship a product, check the share-alike terms of CC BY-SA models yourself, and consider self-hosting the files instead of relying on a third-party CDN.
• Explicit non-clinical scope. The README says ecg-trace is "not for diagnostic use" (its wave detector is a heuristic labelling aid for clean rhythms), dermatome-map is "a simplified illustration, and not a clinical tool", and image-viewer is "not a medical device". Its DICOM reader handles only uncompressed single-frame grayscale files.
• No bundled WHO data. growth-chart takes reference tables as input. It ships no growth data because WHO's standards need WHO's permission for commercial reuse. A parser helper reads tables in WHO's anthro layout that you supply.
• Accessibility. The tree uses a roving tab stop, separate aria-selected and aria-checked, and accessible names that start with the visible state text (WCAG 2.5.3 "label in name"). The docs say keyboard checks are automated in real consumer apps, but "a screen-reader usability audit is still outside the verified scope".
• Scale limits. The anatomy tree is for small and medium in-memory datasets: no virtualization, lazy loading, multi-select or drag and drop.
• Tooling. Bun monorepo with Turborepo, a Next.js 16 docs site, and a Playwright suite against installed consumer apps. AGENTS.md tells coding agents to leave browser tests to CI because WebGL in software rendering makes a laptop unusable.

Caveats

• Weeks old, a single main author, 0.x versions, and releases every day or two. Expect breaking changes and re-installs.
• Items work only with Base UI shadcn styles. A project on the Radix style needs migration first.
• Third-party model licenses and the runtime CDN fetch are the biggest practical issues for a product team.
• The architecture doc is detailed, but I did not check its claims by running tests. Rendering quality and performance of the 42 organ models are unverified here.
• Medical content is simplified illustration. Do not use it for diagnosis or clinical decision-making.