AIHub

AIHub is a lightweight, open-source platform for discovering AI models and datasets. It doesn't host any model or dataset files itself — everything is hosted externally, primarily in GitHub repositories. AIHub simply reads a pair of index files and turns them into a browsable, searchable site.

The whole site is plain HTML, CSS, and JavaScript. No build tools, no frameworks, no backend, no database. It's designed to run directly on GitHub Pages.

Project structure

AIHub/
│
├── index.html          Homepage
├── models.html          Model listing (search, filter, sort)
├── datasets.html        Dataset listing (search, filter, sort)
├── model.html            Model detail page
├── dataset.html          Dataset detail page
│
├── style.css              Shared styling for every page
├── script.js               Shared data loading, rendering, and page logic
│
├── Indexes/
│   ├── ModelIndex.json    List of all models
│   └── DatasetIndex.json   List of all datasets
│
└── README.md

How it works

models.html and datasets.html fetch Indexes/ModelIndex.json and Indexes/DatasetIndex.json respectively, then render a card per entry. Each card links to a detail page (model.html?id=... or dataset.html?id=...), where id is the model or dataset name converted to a URL-safe slug (lowercased, non-alphanumeric characters replaced with hyphens). The detail page re-fetches the same index, finds the matching entry by slug, and renders its full details.

Because the site is entirely index-driven, adding a new entry to a JSON file is enough to make it show up everywhere: the listing page, search, tag filters, sorting, the homepage stat counters, and its own detail page.

How to add a model

1. Push your model's files to its own GitHub repository (or wherever you want to host the actual files — GitHub Releases is a common choice for the download link).
2. Open Indexes/ModelIndex.json and add a new object to the array:

{
    "name": "My Model",
    "author": "your-username",
    "description": "A short description of what the model does.",
    "downloads": 0,
    "likes": 0,
    "updated": "2026-08-05",
    "tags": ["text-generation", "my-tag"],
    "license": "MIT",
    "github": "https://github.com/your-username/my-model",
    "download": "https://github.com/your-username/my-model/releases"
}

3. Commit and push. The model will appear on models.html and at model.html?id=my-model.

How to add a dataset

Same process, but in Indexes/DatasetIndex.json:

{
    "name": "My Dataset",
    "creator": "your-username",
    "description": "A short description of what the dataset contains.",
    "size": "500MB",
    "samples": 5000,
    "updated": "2026-08-05",
    "tags": ["computer-vision", "my-tag"],
    "license": "MIT",
    "github": "https://github.com/your-username/my-dataset",
    "download": "https://github.com/your-username/my-dataset/releases"
}

How the index files work

Each index is a plain JSON array. There's no size limit built into the format — the site is written to handle anything from a handful of entries to thousands, since search, filtering, and sorting all happen client-side against the loaded array. If AIHub eventually needs to support very large catalogs, the index files can be paginated or split without changing the page structure, as long as script.js is updated to fetch the right pages.

Two fields matter most for how an entry is found and displayed:

• name — used to generate the detail-page URL slug. Keep names unique; two entries that slugify to the same value will collide.
• tags — used to build the tag filter chips on the listing pages.

Hosting on GitHub Pages

1. Push this repository to GitHub.
2. In the repository, go to Settings → Pages.
3. Under Build and deployment, set the source to Deploy from a branch, pick the branch (e.g. main) and the root (/) folder.
4. Save. GitHub Pages will publish the site at https://kerbalmissile.github.io/AIHub/.

No further configuration is needed — every path in the site is relative, so it works whether it's hosted at the root of a domain or in a sub-path like a GitHub Pages project site.