A personal "signal over noise" digest that pulls stories from RSS feeds, scores them against your interests, tracks a small price watchlist, and emails a compact 3-day brief via Mailgun.
- Built for automation: run on a schedule, keep lightweight state, and get the digest in your inbox.
- Built for personalization: relevance scoring via topics/regions/keywords + recency boost.
- Built for extensibility: a clean pipeline you can swap pieces in/out (better scoring, LLM summaries, real price APIs).
Add your images to docs/screenshots/ and replace these placeholders.
- Ingest RSS feeds into a rolling cache (
/data/state.json). - Score items for relevance (topics/regions/keywords + recency) and select the top stories.
- Detect "developing stories" by comparing title tokens in the last N days vs the previous window.
- Fetch prices for a watchlist (currently a naive HTML scan; designed to be replaced with APIs).
- Compose a Markdown digest and send it via Mailgun.
- Runtime: .NET 10 (
net10.0), C# - Integrations: Mailgun HTTP API
- Content ingestion:
CodeHollow.FeedReader(RSS) - HTML parsing/scraping:
AngleSharp(price fetcher) - Config:
DotNetEnv(.envfor local dev) - Containerization: Docker
- CI: GitHub Actions (.NET build workflow)
- Optional/experimental: OpenAI .NET SDK (see
NewsAi.cs)
## Developing stories (what changed)
- **ai chips** - now: 9, prev: 2, delta: +7
## Worldwide but relevant to you
- [Headline...](https://...)
## Price trends (watchlist)
- **Example Product** - HOLD. Price above recent lows/avg. Latest: 123.45 USD on 2025-01-01Prereqs: .NET 10 SDK, Mailgun API key + domain.
cd Cosmic-Digest
dotnet restoreCreate your .env:
- PowerShell:
Copy-Item .env.example .env - macOS/Linux:
cp .env.example .env
Run:
dotnet run --project ai-newsletter.csprojNote: state persists to /data/state.json. On Windows, this maps to C:\\data\\state.json (root of the current drive). For a cleaner setup, prefer Docker (below) or change the path in StateStore.cs.
Set these in .env (see .env.example):
| Variable | What it does |
|---|---|
MAILGUN_API_KEY |
Mailgun API key |
MAILGUN_DOMAIN |
Mailgun domain (sandbox or verified) |
MAIL_TO |
Recipient email |
MAIL_FROM |
Sender email (defaults to bot@{MAILGUN_DOMAIN}) |
RSS_FEEDS |
Comma-separated RSS feed URLs |
PREF_TOPICS |
CSV topics you care about |
PREF_REGIONS |
CSV regions you care about |
PREF_KEYWORDS |
CSV high-signal keywords |
PRICE_WATCH |
`name |
OPENAI_API_KEY |
Only needed if you wire in LLM summaries |
cd Cosmic-Digest
docker build -t cosmic-digest .
docker run --rm --env-file .env -v cosmic-digest-data:/data cosmic-digest- Pipeline-style design: each stage is a small, testable unit (ingestion -> scoring -> trends -> prices -> compose -> send).
- State management: rolling cache for news + time-series price history in a single JSON file.
- Trend detection: windowed comparison to surface "what changed" instead of just "what happened".
- Wire in LLM summaries for the selected headlines (optional mode)
- Add scheduling (cron / hosted job)
- Replace naive price scraping with API-backed integrations
- Add web dashboard for configuration and history
- Swap JSON file state for a database when needed
- Never commit
.env(API keys). Use.env.exampleas a template. - For production, prefer platform secrets (GitHub Actions/Fly/your host).
MIT