Argus

AI safety & evaluation research daily digest — delivered to your inbox every morning.

Argus monitors arXiv, Semantic Scholar, HuggingFace Papers, Reddit, and researcher blogs for the latest work on AI safety, alignment, evaluations, interpretability, and governance. It filters for relevance with Claude Haiku, generates a structured digest with Claude Sonnet, and delivers it via email + a committed Markdown file.

---

Quickstart (5 minutes)

1. Fork this repo on GitHub
2. Go to Settings → Secrets and variables → Actions and add your secrets (see table below)
3. Go to Actions → Daily Digest → Run workflow to trigger your first digest
4. Check the digests/ folder for your output

That's it. GitHub Actions runs the pipeline every day at 7am UTC automatically.

---

Required Secrets

Secret	Required	Where to get it
ANTHROPIC_API_KEY	Yes	console.anthropic.com
RESEND_API_KEY	Email	resend.com (free tier: 3000/mo)
DIGEST_EMAIL_TO	Email	Comma-separated list of recipient addresses
REDDIT_CLIENT_ID	Reddit	reddit.com/prefs/apps
REDDIT_CLIENT_SECRET	Reddit	Same app page
SLACK_WEBHOOK_URL	Slack	Slack app incoming webhook

Minimum viable setup: only ANTHROPIC_API_KEY. The digest is committed to the digests/ folder even without email.

---

Estimated Cost

Without Twitter	~$1–2/month
Claude Haiku filtering	~$0.003/day
Claude Sonnet summarization	~$0.05/day
Email (Resend free tier)	Free

---

Changing the Schedule

Edit .github/workflows/daily_digest.yml:

- cron: "0 7 * * *"   # 7am UTC (default)
# - cron: "0 13 * * *"  # 8am EST
# - cron: "0 17 * * *"  # 9am PST

---

Local Development

git clone https://github.com/your-username/argus
cd argus
cp .env.example .env   # fill in your API keys
pip install -e ".[all]"

make fetch     # test fetchers only
make digest    # run full pipeline locally (no git push, no email)
make test      # run test suite

---

Adding a Source

1. Create argus/fetchers/my_source_fetcher.py extending BaseFetcher
2. Implement async def fetch(self) -> list[RawItem]
3. Add it to _fetch_all() in argus/runner.py

---

Customizing the Digest

• Filter threshold: change filter.threshold in config/sentinel.yaml
• RSS feeds: add/remove entries in config/rss_feeds.yaml
• Digest template: edit templates/digest.md.jinja2
• Summarizer prompt: edit SYSTEM_PROMPT in argus/pipeline/summarizer.py

---