TL;DR
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
HackerTube is a YouTube-style site assembled from videos recommended in Hacker News comments. Its creator says the project filtered about 40,000 comments into 6,454 videos across 150 topics, and built the site in roughly two hours with AI assistance.
Developer Andrej Karpathy says he has published HackerTube, a YouTube-style site built from videos recommended in Hacker News comments between 2007 and 2026. The project organizes 6,454 videos across 150 topics, offering an alternative way to browse YouTube without relying on its standard recommendations.
In a post dated October 31, 2026, Karpathy said the project began after he found YouTube’s recommendations and thumbnails unappealing. He asked the AI coding model he called Opus 5.5 to help build a clone centered on videos recommended by the Hacker News community.
According to Karpathy, the workflow gathered Hacker News comments, filtered them, and used another model, Sonnet, to help remove spam and lower-quality material. The resulting dataset contains about 40,000 comments that recommend YouTube videos, which the project reduced to 6,454 distinct videos. The post does not describe an independent audit of those figures or the filtering process.
Karpathy says HackerTube presents the videos in a familiar YouTube-style interface, uses a frame from the middle of each video instead of its uploaded thumbnail, and generates personalized recommendations without user accounts. He says recommendation state remains in the browser’s local session. The site also offers an export of its data for people who want to create their own version.
A Different Route to Video Discovery
HackerTube tests a simple alternative to recommendations selected from a viewer’s YouTube history: a catalog based on links that people posted in Hacker News discussions. That changes the source of discovery, not the underlying video host; the project’s videos are still YouTube videos, according to its description.
The approach may appeal to viewers who want recommendations shaped by a particular online community rather than by their personal viewing history. Its no-account design also means the creator says it does not require a user profile or store recommendation state on a server. The trade-off is that the source community and its posting habits determine what appears: a Hacker News-derived catalog is not a representative survey of YouTube or of viewers generally.
The project also offers a small example of how AI-assisted coding and text processing can be used to assemble a working media interface from a large discussion archive. Karpathy’s account is a first-person description, however, and does not establish how well the site performs, how diverse its catalog is, or whether its recommendations are useful to people beyond its intended audience.
Products Worth Considering
DJI Mini 4K Fly More Combo
Sonos Era 100 SL Compact, Microphone-Free Speaker – Black
Logitech MX Keys S Wireless Keyboard, Graphite
As an affiliate, we earn on qualifying purchases.
From HN Comments to Catalog
Hacker News is a discussion forum where users share and discuss technology-related links. Video recommendations in its comment threads can include talks, tutorials, documentaries, and other material, but the source material does not provide a breakdown of HackerTube’s 150 topics or the number of videos in each one.
Karpathy described the site as a project completed in about two hours, including building the interface, deploying it to a virtual private server, and addressing security and performance. That duration is his estimate of the build process, not a third-party assessment of the site’s security or reliability. The post credits Opus 5.5 and Sonnet with parts of the work, but does not specify which tasks were completed by the models versus by Karpathy.
Filtering and Coverage Questions
The post does not publish the dataset, a list of included videos, or detailed criteria for deciding which comments and recommendations counted. It is therefore unclear how the project handles duplicate links, deleted videos, or recommendations that were sarcastic or incidental. The reported 40,000 comments and 6,454 videos are figures supplied by Karpathy, not independently verified counts in the material provided.
Other details are also unspecified: how the site’s personalized recommendations are generated, whether its topic labels were checked by people, and how much the catalog changes over time. Karpathy says recommendation state stays in a local browser session, but the post does not provide a technical privacy review or explain what information the hosted site may otherwise collect.
Testing the Public Site
The next step for readers is to try the site and evaluate the catalog, interface, and recommendation behavior directly. Karpathy says users can export the site’s data to make their own version, but the post does not announce a release schedule, maintenance plan, or plans to update the dataset.
Whether HackerTube becomes a continuing service or remains a small demonstration is not stated. Its value will depend on how useful visitors find a community-sourced catalog and whether the site’s data and filtering choices are made clear enough for users to assess.
Key Questions
What is HackerTube?
It is a YouTube-style website populated with videos recommended in Hacker News comments, rather than a new video-hosting platform.
How many videos does it include?
Karpathy reports 6,454 videos drawn from about 40,000 comments, spanning 150 topics and recommendations posted from 2007 through 2026.
Does HackerTube require an account?
Karpathy says it works without user accounts and keeps recommendation state in the browser’s local session. The source post does not provide an independent privacy assessment.
How were the videos selected?
Karpathy says Hacker News comments were collected and filtered, with Sonnet used in a workflow to remove spam and lower-quality material. He has not provided detailed selection rules or an independent audit in the source material.
Can users make their own version?
Karpathy says the site offers a data export intended to help users create their own version. The post does not detail licensing or other reuse terms.
Source: karpathy.bearblog.dev
