A small CLI that builds a daily digest of credible agentic AI / LLM news for an AI product manager working in a regulated corporate environment.
It pulls from a curated mix of RSS feeds (Simon Willison, Anthropic, Hugging Face, Techmeme, Latent Space, Import AI, Ethan Mollick, Interconnects, AI Snake Oil, Pragmatic Engineer) plus Hacker News stories matching agentic-AI queries, scores everything for relevance + recency + source trust, dedupes, and either prints the ranked list or asks Claude to synthesise a themed brief.
Live page: https://raidianblaster.github.io/Content-Finder/
cd Content-Finder
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
# optional, for Claude-synthesised digests
export ANTHROPIC_API_KEY=sk-ant-...Plain ranked list (no API key needed):
python3 content_finder.py --no-summarize --days 2Claude-synthesised brief (requires ANTHROPIC_API_KEY):
python3 content_finder.py --days 2Save to a file:
python3 content_finder.py --days 2 --out digest-$(date +%F).md- Top story — single most important development.
- Models & capability releases
- Agentic engineering & tooling
- Enterprise, regulation & governance
- Worth a deeper read
Sections with no relevant items are skipped rather than padded.
Edit sources.yml to change RSS feeds, Hacker News queries, keyword weights,
and per-source trust bonuses. The loader validates HTTPS feed URLs, duplicate
source names, trust ranges, and required top-level keys.
ROADMAP.md is the source of truth. Milestone 0.1 tracing, 0.2 score-feature
logging, and 0.3 CI/do-no-harm gate are landed in code. The current trunk path is:
fix the quarantined archive test, build the eval harness, then train the first
self-tuning scorer from feedback labels.
Add to cron for a 7am local digest:
0 7 * * * cd ~/00-Claudecode-projects/Content-Finder && \
.venv/bin/python content_finder.py --days 1 \
--out ~/Documents/ai-digest-$(date +\%F).md