Multi-Source Tech News Digest

Most "tech news" feeds are 90% noise. Aggregating them with simple scoring — your taste, your stack, your interests — turns a 200-item firehose into a 10-item daily reading list. Flowly's web tools and memory let you express "what I care about" once and have it applied across every source.

What it does

Pulls from Hacker News, GitHub trending, ArXiv (CS), and configurable RSS
Scores each item against your stated interests (memory-stored profile)
Deduplicates across sources (same story covered in 3 places = 1 entry)
Delivers top 10 daily as a digest with one-line summaries + your-rating
Optional: "I want to read this later" → saves to memory tagged #read-later

What you'll need

Web fetch tool — built in
A list of RSS feeds you trust
Your interests profile in memory
Cron

Setup

1. Define your reading taste

Send this once:

Send to Flowly

Remember my reading interests: - High signal: AI agents, plugin architectures, LLM evals, Python internals, infrastructure war stories - Mid signal: HN frontpage if 200+ points, GitHub trending repos with 10k+ stars, anything mentioning "we built" + technical depth - Skip: blockchain price news, crypto launches, AI hype with no technical content, generic "X is dying" articles, listicles Tag my interests profile "reading-taste".

2. Configure the RSS list

Send to Flowly

Index these RSS feeds in memory tagged "rss-feeds": - https://news.ycombinator.com/rss - https://github.com/trending.atom - https://arxiv.org/rss/cs.AI - (your favourites here) For each feed, store: url, name, weight (1-10 for trustworthiness).

3. Schedule the digest

Send to Flowly

Set up cron "tech-digest" Monday-Friday at 8 AM: 1. Fetch all feeds from rss-feeds memory 2. Pull HN top 30, GitHub trending top 20, ArXiv CS new 50, plus all RSS items from last 24h 3. Filter against my reading-taste profile — drop hard-skips, keep high signal 4. For each survivor, fetch the article, write a 2-sentence summary + a 1-10 score 5. Deduplicate (compare titles + URLs across sources) 6. Sort by score, take top 10 7. Format as Telegram message: - Title (linked) - Score: 8/10 - 2-sentence why it matters - Source If nothing scores above 6, just send "Quiet day — no high-signal items."

Tips

First week is calibration. The scoring will misfire — too generous, too harsh, wrong topics. Fix it via memory: "Remember: I do NOT care about Rust release notes. I DO care about Rust performance war stories." Scoring sharpens within 5–7 days.
Don't fetch full content for everything. Title + abstract is often enough. Fetching every 50-page PDF blows your tokens.
Avoid score inflation. It's better to deliver 5 items rated 9/10 than 15 items where most are 6. Add a hard floor.
Pair with read-later. Save scored-but-not-today items with #read-later. Sunday cron: "summarise everything I tagged #read-later this week, group by topic, ask which I want to keep."