lijunliu-gh/ai-news-digest：聚合 Anthropic、OpenAI、Google、GitHub 官方 AI 更新的三语资讯站

Official AI product updates from the past three months across GitHub, Anthropic, OpenAI, and Google.

🌐 Live Site: https://lijunliu-gh.github.io/ai-news-digest/

Features

• 📈 Trend Chart — Interactive line chart showing monthly update frequency per company, color-coded and auto-generated from live data
• 🗞️ Official Release Surfaces — Aggregates both marketing news and developer-facing changelogs / release notes from Anthropic, OpenAI, Google, and GitHub
• 🌍 Trilingual — Switch between 中文 / 日本語 / English with data-level localization for every item
• 🌓 Dark & Light Mode — Polished dual themes with gradient backgrounds, glassmorphism cards, category-colored borders, and a sliding toggle switch
• 🔍 Search & Filter — Filter by company, source type (News / Changelog / Release Notes), month, and full-text search across titles, summaries, and tags
• ⬆️ Back to Top & Month Navigator — Floating back-to-top button and a sidebar month indicator for quick navigation through long feeds
• ♻️ Automated Refresh — GitHub Actions refreshes the digest five times a day using only free GitHub-native automation
• 🗄️ Rolling Window + Archive — The homepage keeps only the latest 3 months while older items move to an archive dataset
• 📱 Responsive — Works on desktop, tablet, and mobile (iOS & Android)
• 📡 RSS Feed — Subscribe via feed.xml with any RSS reader; auto-generated on every refresh
• 🧹 Smart Deduplication — Fuzzy title matching merges duplicate entries across News and Changelog sources automatically
• ⚡ No Frontend Build Step — Pure HTML / CSS / JS frontend, with a small Python dependency only for the refresh script

Project Structure

ai-news-digest/
├── index.html          # Main page
├── feed.xml            # RSS 2.0 feed (auto-generated)
├── css/
│   └── style.css       # Styles (dark/light theme, responsive)
├── js/
│   └── app.js          # App logic (i18n, filtering, rendering)
├── data/
│   ├── archive.json    # Older items moved out of the active 3-month window
│   ├── digest.json     # Active news data for the latest 3 months
│   └── translation-cache.json # Cached machine translations for titles and summaries
├── scripts/
│   └── update_digest.py # Official-source fetcher and retention manager
├── .github/
│   └── workflows/
│       └── update-digest.yml
├── CHANGELOG.md
├── CONTRIBUTING.md
├── LICENSE
├── requirements.txt
└── README.md

Data Sources

Adding News

Edit data/digest.json. Each item follows this structure:

{
  "id": "2026-04-08-001",
  "date": "2026-04-08",
  "category": "anthropic",
  "title": {
    "zh": "中文标题",
    "ja": "日本語タイトル",
    "en": "English Title"
  },
  "summary": {
    "zh": "中文摘要",
    "ja": "日本語の要約",
    "en": "English summary"
  },
  "url": "https://...",
  "tags": ["tag1", "tag2"],
  "source": "Anthropic News",
  "sourceType": "news"
}

Categories: github, anthropic, openai, google

Source types: news, changelog, release-notes

Automation

This repository uses a scheduled GitHub Actions workflow to keep the site current.

• Workflow: .github/workflows/update-digest.yml
• Script: scripts/update_digest.py
• Active window: latest 3 calendar months
• Archive target: data/archive.json

Scheduled refreshes run five times a day, evenly spaced ~4 h 48 min apart:

GitHub Actions uses UTC internally; the workflow cron entries match the UTC column.

Update Rules

• Only official product-oriented sources are ingested.
• Official sources include both public news pages and developer release-note / changelog surfaces.
• For Anthropic, the scraper discovers articles via the sitemap.xml and extracts published dates and metadata from each article page, giving full coverage beyond the front-page listing.
• For Google, the digest combines Google Blog, Google Cloud release notes, and Google DeepMind product-news posts (via RSS). Keyword filters match product names (Gemini, Gemma, NotebookLM, AI Studio, etc.) while excluding non-product research and safety content.
• For GitHub, Product News and Changelog entries use separate relevance rules: product posts trust official AI-oriented category labels and full article content, while changelog entries apply stricter keyword matching to avoid platform-maintenance noise.
• Locale fields that still match the English source text are treated as untranslated and regenerated during refreshes.
• The homepage dataset is rebuilt from scratch on every run.
• Automated refreshes fail closed when the rebuilt dataset drops below health thresholds or a required official source disappears.
• Health thresholds are configurable through GitHub Actions environment inputs for manual runs and workflow defaults for scheduled runs.
• Manual GitHub Actions runs can override validation thresholds from the workflow dispatch form without editing the script.
• Items older than 3 months are removed from data/digest.json.
• Removed items are preserved in data/archive.json.
• Existing manual translations are preserved when the same URL already exists.
• Newly fetched English items are machine-translated into Chinese and Japanese, then cached for reuse in later refreshes.
• The translation cache lives in data/translation-cache.json and is committed so future runs reuse the same phrasing.
• The homepage shows whether an item came from a news page, changelog, or release-notes surface.
• Each GitHub Actions run publishes a job summary with total items, per-source counts, translation stats, validation config, and any validation issues.
• Cross-source duplicates (same event reported in both News and Changelog) are automatically merged using fuzzy title matching, preferring the richer News source.
• An RSS 2.0 feed (feed.xml) is regenerated on every refresh with the latest 50 items.

Development

No frontend build tools are needed. For the automation script, install Python dependencies first:

python3 -m pip install -r requirements.txt

Then open index.html in a browser, run a local server, or rebuild the datasets manually:

# Python
python3 -m http.server 8000

# Node.js
npx serve .

# Refresh digest data
python3 scripts/update_digest.py

Contributing

See CONTRIBUTING.md for guidelines.

License

This project is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). See LICENSE for the full text.

Curated by RoundTable AI Lab