HackerTube
HackerTube is a video-discovery website created by Andrej Karpathy with Claude Code. It collects YouTube videos repeatedly discussed on Hacker News and adds searchable descriptions, topic categories and recommendations. Karpathy described the project in a blog post dated October 31, 2026.[1]
The website's November 2026 description lists 6,454 videos, 39,834 labeled Hacker News comments and 150 topics within 16 areas, drawn from comments published between 2007 and 2026. These are the project's reported collection figures, rather than an independent audit of the complete archive.[2]
How the collection was built
According to the project's methodology, its pipeline first scanned a public Hacker News archive for YouTube links. A candidate video needed links from at least three distinct commenters, with substantive commentary, across at least three different months. This filtered out one-off links before the model-labeling stage.[2]
Claude then classified the context of each mention, distinguishing recommendations from citations, jokes and criticism. It used the comment and the title of its parent Hacker News story. The pipeline also produced short quotations and topic labels. Video cards combine YouTube information with the associated discussions to describe a video's subject, format, intended level and continued relevance.[2]
The ranking favors recommendations recurring across separate months rather than a single burst of attention. This makes the selection a particular view of Hacker News discussions, rather than a survey of every YouTube video or a general measure of video quality.[2]
Browsing and search
The application provides several routes into the collection.[2][3]
| Feature | Documented behavior |
|---|---|
| Home | Mixes recurring favorites, lesser-watched selections, recent recommendations, unexplored topics and personalized suggestions |
| Longs | A filtered browsing view for longer videos |
| Search | Semantic search that can match subject matter beyond literal title keywords |
| Categories and playlists | Topic-based navigation and prepared collections |
| Video pages | Descriptions and relevant Hacker News comments, with links to their original threads |
| Liked, History and Watch later | Browser-local lists; likes also influence Home recommendations |
Video cards explain why an item appears, for example as a recurring recommendation or a less-watched selection. The site supports keyboard shortcuts, including / for search, j and k for moving through videos, and ? for the shortcut list.[2]
HackerTube normally shows a frame from the middle of a video in place of its uploader-selected thumbnail. Its Settings page lets the reader switch back to the original thumbnails.[1][4]
Accounts and browser state
HackerTube does not require an account. Its About page says likes, viewing history and the Watch later list are stored in the current browser. Videos use YouTube's privacy-enhanced embed. Browser-local state should therefore be distinguished from a synchronized account shared across devices.[2]
Published data
The project offers a data download for reuse in other interfaces and analyses. Its About page describes the following files.[2]
| File | Contents |
|---|---|
videos.jsonl | Video identifiers, summaries, classifications, topic assignments, selected quotations and Hacker News statistics |
mentions.jsonl | Labeled comments, with identifiers, authors, months, story titles and stance annotations |
taxonomy.json | The topic hierarchy and descriptions |
related.jsonl | Co-recommendation counts for 3,778 videos, without usernames |
embeddings.npy and embedding_ids.json | Search vectors and their corresponding identifiers |
DATA.md | Field descriptions, construction notes, examples and licensing information |
The page states that the data is available under ODC-By 1.0, with attribution to HackerTube. It excludes YouTube's own descriptions, view counts and durations from redistribution, while retaining video identifiers through which users can obtain metadata separately. This description concerns the downloadable dataset; it does not establish that the complete application source code has a public software release.[2]
Development with coding agents
Karpathy reported building the project in about two hours with Claude Opus 5.5. The site's credits attribute its pipeline, search, recommendations and interface to Opus working in Claude Code, with teams of Claude Sonnet 5.5 agents producing comment labels, video cards and the topic list.[1][2]
In his social-media announcement, Karpathy said Claude had filtered 43 million comments from about 20 years of Hacker News. That figure describes the initial archive, rather than the smaller set of labeled video mentions listed on the application.[5]
Karpathy described three manual setup tasks: obtaining a YouTube API key, starting a virtual private server and purchasing the domain. He said the agent subsequently handled deployment, security configuration and optimization. These are his account of the workflow and development time, not an independently measured productivity comparison.[1]
Scope and limitations
The catalog reflects which videos people linked repeatedly on Hacker News and how the model interpreted those comments. Its stance labels and summaries are generated annotations. Original discussion links let readers inspect the underlying context, but recurring recommendations alone do not verify a video's factual claims.[2]
The project identifies itself as independent of YouTube, Google, Hacker News and Y Combinator. It remains a discovery interface for YouTube videos, rather than a separate video-hosting service.[2][3]
Continue reading
- Claude CodeClaude Code is an agentic software development product made by Anthropic. It uses models from the Claude family to inspect software projects, propose and apply changes, run development tools, and report…Mentioned twice above
- Andrej KarpathyAndrej Karpathy (born October 23, 1986) is a Slovak-Canadian computer scientist, AI researcher, educator, and entrepreneur who was a founding member of OpenAI, served as Tesla's Senior Director of AI and head…Linked in the introduction
- Search EngineA search engine is a software system that retrieves information from a corpus (the web, a private dataset, or a document store) and ranks results by relevance to a user query.Also in Information Retrieval
- Recommender SystemA recommender system (also called a recommendation system or recommendation engine) is an information filtering system that predicts the preference a user would give to an item and uses those predictions to…Also in Information Retrieval
- Semantic ScholarSemantic Scholar is a free, AI-powered academic search engine and open research-data platform built by the Allen Institute for AI (AI2, also styled Ai2)Also in Information Retrieval
- TavilyTavily is a search engine and API platform built specifically for AI agents and retrieval-augmented generation (RAG) systems.Also in Information Retrieval
More to read
Topics:AI Code GenerationAI Tools & ProductsInformation Retrieval
References
- ^1 ^2 ^3 ^4Karpathy, Andrej. HackerTube. Creator's blog, October 31, 2026. Project origin and reported development workflow.
- ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12HackerTube. About HackerTube. Methodology, collection statistics, data export, privacy and credits. Read in the rendered application on November 4, 2026.
- ^1 ^2HackerTube. Live application. Navigation, video cards and browsing views. Inspected November 4, 2026.
- ^HackerTube. Settings. Thumbnail preference. Inspected November 4, 2026.
- ^Karpathy, Andrej. HackerTube announcement. Creator-reported archive size and workflow, November 2026.
Improve this article
Add missing citations, update stale details, or write a clearer explanation. Anyone can edit: changes go live when you publish, and every edit is checked by an editor.
1 revision · v1 · 909 words · full history
Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify
Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here
Cite this page: AI Wiki. "HackerTube." aiwiki.ai, updated 4 Nov 2026. CC BY 4.0. https://aiwiki.ai/wiki/hackertube