HN Flash is a daily AI-summarized digest of the top 20 Hacker News stories.
Each story gets a one-sentence hook, a short summary of the linked article, and a paragraph capturing the comment thread's substance: what HN found wrong, the loudest counterpoints, the pro-tips that emerged.
All content originates from Hacker News and the articles it links to. Direct links to originals and to HN discussion are always provided.
A Python script runs once a day on GitHub Actions at 14:00 UTC. It queries the Hacker News Algolia API for stories with at least 100 points from the last 36 hours and takes the top 20 by score. For each story it fetches the article body via r.jina.ai (a free reader-mode proxy), the comment tree via Algolia, and the raw article HTML in parallel. Claude Haiku 4.5 generates the three summary fields via a single prompt.
Author, publisher, and date are extracted from the article's HTML metadata (JSON-LD, Open Graph), never inferred by the LLM.
Stories that appeared in a previous digest reuse their cached summary text and only refresh score and comment counts. That keeps the daily run at roughly $0.05 to $0.20 in API credits depending on how many stories are new, but it means comment summaries for stories that linger on the front page for multiple days are frozen at first-capture time, not current.
Storage is Turso (managed SQLite). The site is pre-rendered Jinja2 templates committed to the repo and served from Cloudflare Pages.
AI summaries miss subtlety, smooth over disagreement, and occasionally get details wrong.
When an article is paywalled, bot-blocked, or rendered entirely by JavaScript, the body fetch returns nothing and the article summary is skipped; the title hook and the comment summary still appear. Reuters, Bloomberg, and a few WSJ-class sites hit this regularly.
Only the highest-ranked comment threads are read; niche-but-valuable reply chains can be missed.
Everything is in English.
Daily at 14:00 UTC (~10:00 ET / 15:00 CET / 23:00 JST). The archive is permanent. Subscribe via RSS.
Anonymous analytics via Umami. No cookies, no personal data, no individual identification.
This is a hobby project, and I prefer to stay incognito. For bug reports and misc reach me at hello@hnflash.com.