issueops-digest

AI-powered news curation using GitHub Issues as your UI. No frontend, no server, no monthly fees.

Search → AI Score → Checkbox Pick → Auto-capture

How it works

1. Search: Brave + Tavily (optional) fetch news and web results for your keywords
2. Score: Gemini AI filters out noise, scores articles by relevance and depth
3. Pick: Results appear as a GitHub Issue with checkboxes — open the GitHub app on your phone, tick what you want
4. Capture: Close the Issue → GitHub Actions auto-fetches full articles as Markdown

The full flow in action

Trigger a search via Telegram (optional) or CLI → get notified when results are ready → pick articles → auto-capture:

GitHub Issue with scored checkbox list — tick what you want, close to capture:

Captured article as clean Markdown:

Quick Start

Option A: CLI Mode (run on your machine)

git clone https://github.com/notoriouslab/issueops-digest.git
cd issueops-digest
pip install -r requirements.txt
cp .env.example .env
# Fill in your API keys in .env
python web_digest.py "your search keywords"

Option B: Zero-Server Mode (GitHub Actions only)

1. Fork this repo (keep it private — your Issues contain personal search data)
2. Go to Settings → Secrets and variables → Actions
3. Add these secrets: BRAVE_API_KEY, GEMINI_API_KEY, GITHUB_TOKEN
4. Optional: TAVILY_API_KEY, TELEGRAM_BOT_TOKEN, TELEGRAM_CHAT_ID
5. Edit config.yaml to set your topics
6. The scheduled search runs daily, or trigger manually via Actions → Scheduled Search → Run workflow

API Keys

Service	Required	Free Tier	Get it
Brave Search	Yes	~1000 calls/mo ($5 credit, requires card)	brave.com/search/api
Gemini	Yes	1500 req/day	aistudio.google.com/apikey
GitHub Token	Yes	Free	github.com/settings/tokens (needs repo scope)
Tavily	No	1000 calls/mo	tavily.com
Jina Reader	No	Free (rate-limited)	No signup needed
Telegram Bot	No	Free	Talk to @BotFather

Telegram Notifications (optional)

If configured, you'll get a ping when:

• New Issues are ready — "go check and pick your articles"
• Capture is complete — "your Markdown files are saved"

Setup: create a bot via @BotFather, get your Chat ID, add both as GitHub Secrets. That's it — no server, no polling, just two one-line curl calls in the Actions workflow.

Not configured? Everything still works, you just won't get push notifications.

Configuration

Edit config.yaml:

# Topics for scheduled search
topics:
  - "AI agent framework"
  - "台積電 供應鏈"

# Weight profile: default, news, research
weight_profile: default

# Your trusted sources (higher = more trusted, max 15)
authority_domains:
  bloomberg.com: 15
  reuters.com: 12

Weight Profiles

Profile	Best for	Behavior
default	General use	Balanced scoring
news	Breaking news	Recency weighted higher
research	Deep research	Evergreen content protected from time decay

python web_digest.py "quantum computing" --profile=research

Security

⚠️ Use a private repo. Your Issues contain search keywords, selected articles, and personal notes.

• API keys: stored in .env (local) or GitHub Secrets (Actions) — never committed
• .env is pre-configured in .gitignore
• The app checks on startup that .env is gitignored and warns you if not

Quota Tracking

In CLI mode, the app tracks your API usage in .usage_stats.json (gitignored) and warns you:

• 80%: yellow warning
• 95%: red alert

Limits are configurable in config.yaml under quotas.

How the scoring works

1. Brave Search fetches news + web results; Tavily (optional) adds advanced search
2. Results are merged via round-robin interleaving so no single source dominates
3. Jina Reader fetches full text from quality platforms (Substack, Medium, etc.)
4. Gemini AI scores each result for relevance (0-10), flags duplicates
5. A fuzzy keyword filter (bilingual, order-independent) penalizes off-topic results
6. Authority bonus from your configured trusted domains
7. Recency scoring with evergreen protection in research mode

License

MIT

Credits

Built with the help of AI. Inspired by the belief that anyone with an idea can build useful tools — no CS degree required.