hackertubeguideHackerTube Data Explained
What the public export represents, how the collection was assembled, and what to check before reuse.
The initial collection
Andrej Karpathy described an initial collection of 6,454 videos across 150 topics. Those videos were surfaced from YouTube links recommended in Hacker News comments. The number describes the collection at the time of his published account; it is not a live counter for the original site. The useful lesson is the source trail: a video appears because it was shared in a discussion, not because this site owns or hosts it.
How it was assembled
Karpathy's account describes filtering Hacker News comments with Claude Code and using Sonnet to identify relevant recommendations. YouTube metadata then helped describe the selected videos. HackerTube's browsing layer turns those inputs into a focused discovery experience. The exact collection process belongs to the original author; this guide describes the published account and does not claim to have recreated his production system.
What the export gives you
The author says site data is available to export. A data export can be a starting point for another directory or analysis tool. It is separate from the full website implementation and does not confer ownership of YouTube videos, thumbnails, metadata or Hacker News comments. Do not describe the export as a packaged copy of the original site.
Before you reuse it
Inspect the downloadable data itself before promising fields or features. Check identifiers, source URLs and any topic labels you plan to use. Account for unavailable videos, duplicates and later changes. Preserve links back to the original video and discussion when your interface makes a claim based on them. Review the current rules for the data and third-party media before publishing.