AI News Digest Bot
From 50+ tabs to one digest — automatically, twice a day
12h
fully automatic cycle, no manual input
2 formats
Telegram digest + email longread
0
duplicate articles — Airtable remembers everything
The problem
Following a topic means opening 10-15 sources every day, scanning headlines, deciding what's worth reading, then synthesizing into something useful. That's 1-2 hours of work for 10 minutes of actual insight. Most people either stop following or drown in tabs.
How it works
Schedule fires every 12h
Parser reads RSS sources from Airtable and fetches each feed
Articles filtered
Only articles from the last 24h pass. Already-seen URLs deduplicated against Airtable history
Full content scraped
ScrapingBee extracts complete article body — not RSS excerpts. Ads and banners stripped
AI analyst ranks
All articles read, grouped by topic, scored 1-10. Marketing content filtered out
AI writer publishes
Top picks become a structured longread: intro, numbered sections, key takeaways, sources
Delivered to Telegram + email
Short digest to Telegram, full longread as GitHub Gist to email — no login required
Two AI agents, two jobs
AI Analyst
Reads all articles. Groups by topic. Ranks each 1-10 by importance. Filters marketing content. Returns structured JSON: which articles matter and why.
AI Writer
Takes the ranked selection. Writes a professional longread in Markdown — intro, numbered sections per story, key points, sources, further reading.
Why two agents
One prompt trying to both analyze and write produces mediocre output. Splitting roles forces the analyst to think before the writer writes.
Structured output
Both agents return JSON via schema parsers. The pipeline doesn't break on unexpected formatting — structure validated before passing forward.
The payoff
1-2 hours → 0 minutes daily
Source-checking and synthesis happens automatically, twice a day. You receive the output — nothing to monitor.
Right format for each context
Telegram for a quick scan during the day. Email longread when you have time to go deep.
Zero duplicates, ever
URL normalization strips UTM params and trailing slashes. Same article from different tracking links still registers as seen.
Any niche, same pipeline
Replace RSS sources in Airtable — everything else works unchanged. Tech, finance, law, medicine.
What's under the hood
n8n — orchestration
Two workflows: parser calls writer via Execute Workflow. Both independent and reusable.
ScrapingBee — content extraction
AI extraction rules target article body only. Blocks ads, cookie banners, and nav before they reach the AI.
Claude + Gemini — analysis and writing
Analyst uses a fast model for structured analysis. Writer uses a capable model for long-form content.
Airtable — memory and sources
Two tables: RSS sources and processed URLs. Both editable without touching the workflow.
Open source. Adapts to any niche.
Both workflow JSONs are open. Replace the RSS sources in Airtable — the rest works without changes. Tech news, finance, law, medicine — same pipeline, different feeds.
RSS monitoring
Reads multiple sources every 12 hours. Only fresh articles from the last 24h pass through.
Two-stage AI
First AI analyst ranks articles by importance. Then AI writer turns the top stories into a structured longread.
Dual delivery
Telegram gets a brief digest with key points. Email gets the full longread with analysis and sources.
Memory via Airtable
Every processed URL is saved. Next run skips everything already seen — no repeats ever.