HackerTube – Curated YouTube Clone Built from Hacker News Recommendations

What HackerTube Is and Why It Matters

HackerTube is a YouTube‑style front‑end that serves 6,454 videos recommended in Hacker News comments from 2007‑2026, built automatically by Claude Opus 5.5 in about two hours. It matters because it demonstrates that a high‑quality video feed can be assembled without the clickbait‑heavy, engagement‑maximising algorithms that dominate mainstream platforms.

How the Data Pipeline Works

The pipeline starts with a full scrape of Hacker News comments, filters for YouTube links, and uses Claude’s Sonnet model to remove spam and low‑grade entries. The resulting set contains 40 K comments that actually recommend a video. Metadata (title, thumbnail, upload date) is fetched via the YouTube Data API, then the collection is wrapped in a minimal YouTube‑like UI and deployed to a VPS.

Scraping: All public comments are downloaded via the HN API.
Filtering: Regex identifies youtube.com and youtu.be URLs.
LLM Curation: Claude Sonnet evaluates each comment, discarding spam, self‑promotion, and low‑quality recommendations.
Metadata Enrichment: A Google Cloud API key is used to pull video titles, channel names, and 50 % thumbnails (the latter replaces the often click‑bait original thumbnail).
Deployment: The static site is built, hardened, and served from hackertube.org on a VPS.

"I use Claude Code (desktop app for Mac). I use Opus 5.5 on Medium effort as my default. I don't actually do anything resembling traditional engineering anymore… I only talk to Claude and have it do everything." – Karpathy (HN comment)

Core Features That Differentiate HackerTube

HackerTube intentionally strips away the noise that plagues YouTube.

Quality‑first curation: Only videos that received explicit positive recommendations from HN commenters appear.
50 % thumbnails: The UI shows a clean, cropped thumbnail instead of the often misleading original.
No user accounts: Recommendations are stored locally in the browser session, eliminating tracking.
Keyboard shortcuts: Navigation is fully keyboard‑driven, catering to power users.
Exportable data: A DATA.md file lets anyone feed the collection into their own agents or rebuild the site.

Community Reception and Suggested Improvements

The Hacker News community overwhelmingly praised the concept while offering concrete enhancements.

Recency sorting: Multiple commenters asked for a “new‑to‑old” or “last 24 h” filter to surface recent tech videos.

"I wish there was a way to sort by recency…" – @mixtureoftakes
Topic channels: Users suggested auto‑generated channel pages and playlist exports for easier navigation.

"Publishing the curated videos as automatically maintained playlists would be valuable." – @twalichiewicz
Non‑JavaScript support: A request was made for a fully functional no‑JS fallback.

"Could it be viewed without JavaScript?" – @forevernoob
Legal clarity: Several comments raised concerns about YouTube’s terms of service and potential copyright issues.

"Is this compliant with Google/YouTube terms?" – @albert_e
Mobile UI tweaks: Minor bugs such as clipped like/dislike buttons were reported.

"On mobile layout the like/dislike button is clipped." – @m1keil

Technical Takeaways for Builders

HackerTube showcases a repeatable workflow for building curated media sites.

Prompt‑driven LLM orchestration: Use Claude (or another capable model) to coordinate scraping, filtering, and deployment steps via natural‑language prompts.
Minimal manual effort: Only three manual actions were required—create a Google API key, spin up a VPS, and register a domain.
Open‑source hand‑off: The entire codebase, including a DATA.md manifest, is available for forking, enabling custom UIs, email digests, or TV apps.
Security first: The VPS is hardened (firewall, HTTPS, minimal services) before exposing the site publicly.

Broader Implications for the Video Ecosystem

HackerTube proves that community‑driven curation can out‑perform algorithmic recommendation for niche audiences. It highlights a growing demand for:

Transparency: Users can see why each video was included (the original HN comment).
Control: No tracking, no ads, and no opaque engagement‑maximising loops.
Extensibility: The exported dataset invites downstream projects—RSS feeds, TV apps, or AI‑driven personal assistants.

If more communities adopt similar pipelines, the dominance of monolithic recommendation engines could erode, giving rise to a federated ecosystem of specialized, high‑quality video curations.

How to Get Started with Your Own Curated Feed

To replicate HackerTube, follow these high‑level steps:

Collect source data (e.g., comments from a forum, Reddit threads, or Slack channels).
Identify media links using regular expressions.
Run an LLM filter to discard spam and low‑quality entries.
Enrich with metadata via the platform’s public API.
Build a lightweight UI (HTML/CSS/JS) that mimics the target platform’s layout.
Deploy securely on a VPS or serverless platform.
Publish the data manifest (DATA.md) for community reuse.

Final Thoughts

HackerTube is a compelling proof‑of‑concept that AI‑augmented pipelines can quickly produce high‑quality, ad‑free video experiences tailored to a community’s tastes. The project’s success rests on transparent curation, minimal friction for users, and an open data model that invites further innovation.