Exgentic Website

The website and HuggingFace Space for the Open Agent Leaderboard.

Project Structure

├── index.html              # Main website (HTML structure)
├── styles.css              # All styles (dark/light themes, responsive)
├── scripts/
│   ├── loading.js          # Loading screen animation
│   ├── leaderboard.js      # Leaderboard table + efficiency chart (shared with HF space)
│   ├── animations.js       # GSAP scroll-triggered animations
│   └── main.js             # Theme toggle, header, FAQ, network graph
├── hf-space/
│   ├── index.html          # Leaderboard-only page for HuggingFace Spaces
│   └── README.md           # HF space metadata
├── data/
│   └── results.csv         # Single source of truth for all evaluation data
├── hf-dataset/
│   ├── README.md           # HF dataset metadata
│   └── data/               # Generated parquet (from data/results.csv)
├── .github/workflows/
│   ├── deploy-hf-space.yml   # Auto-deploys HF space on push to main
│   └── deploy-hf-dataset.yml # Auto-deploys HF dataset on push to main
├── results.csv             # Generated website CSV (from data/results.csv)
├── results-README.md       # Data dictionary
├── LICENSE-DATA.txt         # Data license (CDLA-Permissive-2.0)
├── favicon.png
└── CNAME                   # Custom domain (www.exgentic.ai)

Local Development

python3 -m http.server 8080

Then open http://localhost:8080.

Website

The main website is deployed via GitHub Pages from this repo. It includes the full experience: hero section, interactive leaderboard, efficiency chart, findings, network graph, FAQ, and footer.

Custom domain: www.exgentic.ai (configured via CNAME).

HuggingFace Space

The HF Space shows a focused view with just the leaderboard table and efficiency chart.

It reuses styles.css and scripts/leaderboard.js from the main website. A GitHub Action (.github/workflows/deploy-hf-space.yml) automatically deploys updates when relevant files change on main.

HuggingFace Dataset

The HF Dataset contains detailed evaluation results in Parquet format. Source files live in hf-dataset/ and are auto-deployed via .github/workflows/deploy-hf-dataset.yml.

Setup

Add a HF_TOKEN secret to the GitHub repo with a HuggingFace token that has write access to the open-agent-leaderboard/leaderboard space and the open-agent-leaderboard/results dataset.

Updating Data

Edit data/results.csv — this is the single source of truth for all evaluation results. Then run:

python3 scripts/build_data.py

This generates:

• results.csv — website-format CSV (used by the leaderboard and HF space)
• hf-dataset/data/train-00000-of-00001.parquet — parquet for the HF dataset

The GitHub Actions run this build step automatically before deploying.

Resources

• Paper (arXiv)
• Evaluation Framework (GitHub)