⟡ دایره
_Dayereh

A Persian-language news aggregator that costs nothing to run.

No server. No database. No hosting bill — just a static site rebuilt by a cron job.

Overview · Features · How it works · Stack · Quickstart · Deployment

◌ What is Dayereh?

دایره به معنای "دایره" یا "چرخه" است — تجمیع، پالایش و خلاصه‌سازی هوشمند اخبار فارسی، بدون هیچ زیرساخت سرور.

Dayereh pulls news from Telegram channels, RSS feeds, and financial market sources, cleans and structures it, and renders it into a single fast, RTL-native static site. A GitHub Actions cron job does the rebuilding — the entire output is static HTML/XML/JSON, committed straight back to the repo. No backend, no database, no cost.

✦ Features

📡 Multi-source aggregation Telegram public channels · RSS outlets · TGJU · Live Rates 🧠 AI daily bulletin Gemini-generated geopolitical/economic summary with a persistent rolling memory — each bulletin builds on the last instead of starting cold 🛡️ Resilient scraping Exponential backoff, pooled connections sized for worst-case concurrency, Google News proxy fallback for bot-protected domains

🎞️ Full media pipeline Thread-safe download cache, gallery/album support, automatic image & video compression, video thumbnail extraction 🎨 Modern static frontend dark theme · fully RTL ⚡ Zero-cost deploy Static assets only — served straight from GitHub, rebuilt on a cron schedule

⟳ How it works

sources.json
     │
     ▼
  Fetcher  ──►  Parsers  ──►  FeedItem  ──►  Builders  ──►  index.html
(requests/       (Telegram,    (dataclass)    (XML / JSON /
 Selenium)        RSS, TGJU,                    HTML)
                  LiveRates)
                                  │
                                  ▼
                         TextCompressor ──► Gemini ──► news-summary.json
                                                   │
                                                   ▼
                                          memory.md (rolling context)

Step	What happens
1. Fetch	Sources pulled in parallel via ThreadPoolExecutor
2. Parse	Per-source parser normalizes content into FeedItems
3. Media	Thread-safe cache downloads + compresses images/video via ffmpeg/Pillow
4. Build	Per-source XML (MRSS) and a combined JSON price feed
5. Summarize	Feed text is TF-IDF compressed, then sent to Gemini with a fixed analyst persona + rolling memory
6. Render	Static index.html with related-article clustering and read-state tracking
7. Deploy	GitHub Actions commits output back to main on a cron schedule

⚙ Tech stack

🗂 Project structure

.
├── main.py                      # pipeline entrypoint
├── config/
│   └── sources.json             # telegram / rss / tgju / live-rates definitions
├── core/
│   ├── parsers/
│   │   ├── telegram.py          # public channel scraper (posts, albums, polls, video)
│   │   ├── rss.py               # RSS/Atom item parser
│   │   ├── website.py           # generic CSS-selector scraper
│   │   ├── tgju.py              # currency/gold/coin market data
│   │   ├── live_rates.py        # FX/crypto rate feed
│   │   └── tgstat.py            # TGStat channel stats
│   ├── net/
│   │   ├── fetcher.py           # requests + Selenium fetch layer
│   │   └── media_downloader.py  # retrying media downloader
│   ├── converters/
│   │   ├── media_type.py
│   │   ├── media_compressor.py  # ffmpeg/Pillow compression
│   │   └── markdown_to_html.py
│   ├── builders/
│   │   ├── xml_builder.py       # per-source MRSS output
│   │   ├── json_builder.py      # price feed JSON
│   │   └── html_builder.py      # static site renderer
│   ├── ai/
│   │   ├── gemini.py            # Gemini API client
│   │   ├── text_compressor.py   # TF-IDF summarization pre-pass
│   │   ├── soul.md              # analyst persona / system prompt
│   │   └── memory.md            # rolling summarization memory
│   ├── models/
│   │   └── feed_item.py
│   └── utils/
│       ├── string_utils.py
│       ├── time_utils.py        # Tehran (UTC+3:30) time normalization
│       ├── path_utils.py
│       └── id_generator.py
└── feeds/                       # generated output (XML, JSON, static site)
    └── view/
        ├── index.html
        └── media/

🔧 Configuration

Sources live in config/sources.json:

{
  "telegram": [
    { "app_name": "tgm-example", "title": "کانال نمونه", "url": "https://t.me/s/example" }
  ],
  "rss": [
    { "app_name": "rss-example", "title": "سایت نمونه", "url": "https://example.com/rss" }
  ],
  "tgju": [
    { "slang": "price_dollar_rl", "title": "دلار آمریکا" }
  ],
  "live-rates": [
    { "currency": "GOLD", "title": "طلای جهانی" }
  ]
}

The Gemini key is passed as a CLI argument, never stored in config.

🚀 Quickstart

git clone https://github.com/ch3fferson/dayereh.git
cd dayereh
pip install -r requirements.txt

python main.py                    # build without AI summary
python main.py $GEMINI_API_KEY    # build with AI summary

Output lands in feeds/ and feeds/view/index.html. Skipping the Gemini key reuses the most recent cached bulletin.

📦 Deployment

Deployment is entirely handled by a scheduled GitHub Actions workflow:

• Cron trigger fixed to Tehran time (UTC+3:30, no DST): 30 */2 * * * (UTC)
• Runs on the standard 2-vCPU / 7GB Actions runner
• Commits regenerated feeds/ output back to main

📄 License

MIT — see LICENSE.

تقدیم به همه جاویدنامان ایران