TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
HackerTube is a YouTube-style site built from videos recommended in Hacker News comments. Its creator says an AI-assisted workflow filtered more than 40,000 comments to identify 6,454 videos across 150 topics; details about the filtering process and the site’s audience remain limited.
Developer Andrej Karpathy says he built and launched HackerTube, a YouTube-style video site assembled from recommendations found in Hacker News comments. The project turns an archive of 6,454 YouTube videos, identified through more than 40,000 comments dating from 2007 to 2026, into a browsable feed organized across 150 topics.
In a blog post dated October 31, 2026, Karpathy described starting the project after finding YouTube’s recommendations and thumbnails unsuited to his tastes. He said he asked an AI coding assistant, which he named as Opus 5.5, to help create a video site based on links recommended by Hacker News users. That account is Karpathy’s description of the project’s development, not an independently documented audit of the system.
Karpathy said the workflow downloaded Hacker News comments, filtered them, and used a model he called Sonnet to help remove spam and what he described as low-quality material. The process narrowed the source material to about 40,000 comments recommending videos, yielding 6,454 distinct videos across 150 topics. The post does not provide a detailed methodology for how recommendations or topics were classified.
According to Karpathy, the site presents the videos in a familiar YouTube-style interface, uses a frame from the middle of a video rather than its uploaded thumbnail, and supplies custom recommendations without requiring accounts. He also said browsing state remains in the local browser session and that users can export the site’s data to use with their own AI agent. He estimated that recreating the project took about two hours.
A Different Route to Video Discovery
HackerTube offers a small-scale alternative to a feed chosen by a platform’s recommendation system: its starting point is what people in Hacker News discussions linked to over time. That makes the project an experiment in using a community’s public recommendations as a discovery signal, rather than relying solely on each viewer’s watch history or a platform’s personalized feed.
The approach may appeal to viewers who want to browse technical talks, explainers, and other material that has been shared in a developer-oriented community. But the site’s catalog is not a neutral or comprehensive sample of YouTube. It reflects which videos Hacker News users chose to mention, and the filtering process adds another layer of editorial or model-based selection. Its usefulness will depend on the range and quality of those recommendations, as well as how the interface ranks them.
The project also illustrates how an individual can combine an existing public discussion archive, AI-assisted data processing, and a web interface to produce a specialized discovery tool. Karpathy’s account suggests that the work was quick, but does not establish that the same time or results would apply to other projects or users.
Products Worth Considering
Bose QuietComfort Ultra Wireless Noise-Cancelling Headphones with Spatial Audio, Black
Bose QuietComfort Wireless Noise-Cancelling Headphones, Petal Pink
Samsung 49-Inch Odyssey G93SC QD-OLED Curved Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Hacker News Comments to Catalog
Hacker News is a technology-focused discussion site where users submit links and comment on them. Video recommendations appear within those discussions, but they were not originally presented as a dedicated video library. HackerTube’s stated premise is to collect that scattered material and make it browseable in one place.
Karpathy said the source period runs from 2007 through 2026. His post describes using AI tools to filter comments and build the site, but it does not publish the underlying dataset, a full list of processing rules, or a comparison with YouTube’s own recommendation system. The figures of 40,000 comments, 6,454 videos, and 150 topics are reported by the project’s creator.
The site’s stated features include a no-account browsing experience, session-based local state, alternative thumbnails, and an export of its data. These details describe the product as presented in the post; independent testing or information about how long the site will be maintained was not included in the source material.
What the Filtering Leaves Out
The source material does not explain how the workflow decided that a comment recommended a video, how duplicate links or deleted videos were handled, or what criteria were used to remove spam and low-grade material. It also does not specify whether the 6,454 figure represents unique videos after all forms of duplication were checked. The numbers should therefore be understood as figures reported by Karpathy, not independently verified counts.
It is also unclear how the 150 topics were defined, whether users can inspect or challenge classifications, and how the site generates its custom recommendations. The post gives no usage statistics, independent assessment of video quality, or details about the catalog’s availability over time. It does not establish whether the project has any relationship with YouTube beyond presenting YouTube videos through its own interface.
The Site’s Next Test Is Use
The immediate next step is for users to try the site and judge whether the Hacker News-derived catalog makes it easier to find videos they want to watch. Karpathy’s post points to an export feature that could let users work with the collected data outside the site, although it does not describe a longer-term release plan or a schedule for updates.
More information would clarify how the comment-filtering workflow works, how often the dataset changes, and whether the project will remain available. Until those details are published, HackerTube is best understood as a creator-described experiment in community-sourced video discovery rather than a verified measure of video quality or a demonstrated replacement for YouTube’s recommendations.
Key Questions
What is HackerTube?
HackerTube is a YouTube-style video site that, according to its creator, gathers videos recommended in Hacker News comments and organizes them into a browsable catalog.
How many videos does it include?
Andrej Karpathy says the catalog contains 6,454 videos, drawn from recommendations identified in about 40,000 Hacker News comments. Those counts come from his blog post and have not been independently verified in the supplied material.
Does HackerTube require an account?
Karpathy says the site works without user accounts and stores browsing state in the local browser session. The post does not describe every data-handling detail.
How were the videos selected?
Karpathy says an AI-assisted workflow filtered Hacker News comments, removed spam and lower-quality material, and identified video recommendations. The post does not publish the full filtering rules or a detailed account of how topics were assigned.
Is HackerTube affiliated with YouTube?
The source describes HackerTube as an independent site presenting YouTube videos through its own interface. It does not report an affiliation with YouTube.
Source: karpathy.bearblog.dev
