HackerTube

What is it

HackerTube is a curated video discovery site that pulls the YouTube links Hacker News users recommend in comment threads and re-presents them as a browsable, YouTube-like feed. Andrej Karpathy built it and showed it off via a Show HN post on November 2, 2026. Technically it's simple: a scraper that walks HN comments, extracts YouTube URLs, deduplicates them, scores them by points and discussion depth, and serves a clean video grid. There is no upload, no transcoding, no creator program — it's pure curation on top of somebody else's video hosting.

The business significance is bigger than the artifact. HackerTube is a proof of concept for "community-signal curation" — taking a high-trust forum's implicit recommendations and turning them into a structured, searchable product. The same pattern immediately spawned HackerBooks (HN-recommended books), and the wave of copycats suggests the underlying asset isn't video at all. It's the HN recommendation graph, which has been sitting unmonetized for 15 years. For indie developers, that's the real opportunity: whoever builds the durable index and the API underneath these one-off sites owns a defensible layer that Karpathy's weekend project deliberately does not.

Why now

Three things converged in late 2026 to make this land. First, HN's own scale crossed a threshold where comment-thread curation became more valuable than the front page. The site now regularly sees 5,000+ comments on major threads, and the "best video I've seen on X" recommendations buried in those threads were effectively lost — no search, no archive, no ranking. Second, LLM-based extraction got cheap enough that parsing messy comment text for video links, titles, and context is a weekend job instead of a two-week NLP project. Karpathy's build is itself evidence: a single developer shipped a usable product fast.

Third, and most important, there's a demonstrated appetite for "anti-algorithm" discovery. YouTube's recommendation engine is widely disliked among technical users, and HN readers explicitly want recommendations filtered by people who share their taste. The rapid spawn of HackerBooks and similar curation sites within days of HackerTube's launch is the clearest signal: the demand isn't for HackerTube specifically, it's for the category. That window — where the pattern is validated but no incumbent owns it — is exactly when indie products win. In 12 months this niche will either be consolidated by one player or absorbed into HN-adjacent tooling.

Market Evidence

The evidence base is thin but directionally strong. Two independent sources (Show HN and Hacker News discussion) generated 2 tracked mentions with a 100% growth rate and a trend score of 69/100 — high for a nascent term. The stage is correctly labeled nascent: this is week-one territory, not a mature market.

What matters more than raw mention counts is the shape of the signal. A single Show HN post by Karpathy is not a normal launch — it's a distribution event. Karpathy's posts routinely hit the HN front page within minutes, and the resulting comment thread functions as a market survey. The fact that HackerBooks appeared almost immediately is the strongest evidence: copycats don't spawn in days unless builders see an obvious gap. That's competitive validation, not demand validation, but it's the closest proxy available this early.

My read: this is real demand with a fragile timeline. The demand for curated technical video discovery is durable — HN users have wanted it for years. The demand for HackerTube specifically is near zero, because the product is trivially cloneable and its creator has no commercial intent. Treat the signal as "category is hot, no owner yet," not "this product will win." The 0/100 opportunity and demand scores in the dataset should be read as "insufficient data," not "no opportunity" — the scoring model hasn't seen enough volume to calibrate.

Who's Behind It

The central figure is Andrej Karpathy — former Tesla AI director, OpenAI founding member, and one of the most-followed technical voices on the internet. His involvement is both the launch engine and the ceiling: he built HackerTube as a demonstration, not a company, and has shown no interest in monetizing it. That's good news for indie developers. The whale here is the community, not the founder.

The real power centers are Hacker News itself (operated by Y Combinator) and the adjacent builder community on X and in HN comment threads. YC has historically ignored third-party HN-derived products — Algolia's HN Search has operated for over a decade without friction — which suggests tolerance rather than partnership. The copycat wave (HackerBooks and unnamed others) consists of anonymous indie builders, none of whom have shipped a durable product yet. There is no incumbent, no funded competitor, and no platform owner actively defending the space. That's an unusually clean competitive field for a validated category, and it won't stay that way past Q1 2027.

TAM & Market Size

The addressable market is small but high-value. HN's active commenter base is roughly 100,000–200,000 people monthly; the broader "technical professional who distrusts recommendation algorithms" segment is perhaps 2–5 million globally. That's a niche, not a mass market — and that's fine, because niches with high willingness to pay beat mass markets with zero.

Who pays? Three groups. First, individual developers and researchers who want a better discovery feed — realistic price tolerance is $5–10/month, comparable to what this audience pays for Readwise ($9.99/mo) or Setapp. Second, teams doing technical research or content curation — $20–50/seat/month is defensible if you add search, tagging, and export. Third, API consumers building their own curation products — usage-based pricing at $0.01–0.05 per query or $99–299/month for volume tiers.

The dataset's 0/100 opportunity and demand scores reflect insufficient signal, not a dead market. My estimate: a focused product can realistically reach 500–2,000 paying users within 18 months, implying $30K–$200K ARR. That's a lifestyle business, not a venture outcome — which is exactly the right frame for an indie developer. Don't build for TAM; build for a defensible 1,000 true fans who will pay $8/month forever.

Competitive Landscape

The competitive set is thin and mostly weak. HackerTube itself is the category leader by default and has no monetization, no roadmap, and no moat — it's a static site that could be rebuilt in a weekend. HackerBooks is a sibling project with the same limitations. Algolia HN Search is the closest durable incumbent, but it indexes text, not video recommendations, and has no curation layer. YouTube itself is the underlying platform — not a competitor, but a dependency risk.

The gap is obvious: nobody owns the index. Every one of these sites re-scrapes HN on every request, which means no historical ranking, no deduplication across threads, no quality scoring over time, and no API. A product that maintains a persistent, deduplicated, time-weighted recommendation graph — with search, filtering by topic, and export — is meaningfully better than all of them, and the data asset compounds daily.

If Big Tech enters, the realistic candidates are YC itself (unlikely — it doesn't ship consumer products), YouTube/Google (very unlikely — this niche is too small and hostile to their algorithm), or a funded startup noticing the trend. My estimate: you have 3–6 months before a well-funded clone appears. Competition score of 0/100 is accurate today and will not stay accurate. Speed and data ownership are the only defenses.

Business Model

Recommendation: freemium SaaS with an API tier, not one-time purchase. This audience is allergic to subscriptions in principle but pays them readily for tools they use daily (see: Kagi, Readwise, Obsidian Sync). The product needs ongoing scraping and ranking infrastructure, which justifies recurring revenue and aligns incentives.

Pricing structure:

• Free: 50 video recommendations/month, basic search, no export. This is the acquisition channel and the SEO surface.
• Pro — $8/month or $72/year: unlimited recommendations, full search and filtering, saved collections, CSV/JSON export, email digest. Anchored just under Readwise's $9.99 and above "impulse" territory.
• Team — $29/month for 5 seats: shared collections, API access at 1,000 calls/month, Slack integration.
• API — usage-based: $0.02/query above the Team allotment, with a $99/month floor for commercial use.

12-month forecast: conservative $8K ARR (150 Pro users), base $30K ARR (400 Pro + 20 Team + 5 API), optimistic $75K ARR (1,000 Pro + 60 Team + 15 API). CAC estimate: $15–40 via HN launches, technical blog posts, and newsletter sponsorships in adjacent niches (TLDR, Pointer). Payback period: 2–5 months on Pro, under 2 months on Team. The economics work because the audience is concentrated and reachable — you don't need paid acquisition to find the first 500 customers.

MVP Blueprint

Scope ruthlessly. The MVP is the persistent index plus a search UI — nothing else.

Core features (build these):

1. HN scraper that walks comment threads, extracts YouTube URLs with surrounding context, and stores them in Postgres with thread ID, timestamp, and comment score.
2. Deduplication and ranking: time-weighted score combining HN points, comment depth, and recency. Re-rank nightly.
3. Web UI: searchable grid with topic tags, duration filter, and "why this was recommended" context snippet.
4. Email digest: weekly top 20, one-click unsubscribe.
5. Stripe checkout for Pro tier and a simple API key system.

Cut entirely: user accounts beyond email, social features, comments, playlists, mobile app, recommendations-for-you ML, creator tools.

Stack: Next.js on Vercel, Postgres on Supabase, a Python scraper on a $5 VPS with cron, Stripe for billing, Resend for email. Total monthly cost under $50 until you have real traffic.

Fastest path: ship the scraper and a read-only UI in 3 days, post to Show HN with a genuinely useful angle ("search every video HN has ever recommended"), collect emails for the digest, and gate search depth behind signup. Dev estimate: 5–7 days for a solo developer working evenings. The API ships in week 3, after you see what people actually query.

Commercial Opportunities

1. The persistent recommendation index + API. Target: developers building curation tools and researchers studying technical discourse. Revenue: $2K–8K/month from API tiers and Team plans. This beats alternatives because the data asset compounds — every day of scraping makes the index more valuable and harder to replicate, and nobody else is maintaining one.

2. Curated video newsletters for technical niches. Target: engineering managers and senior developers who want a filtered feed without visiting HN. Revenue: $1K–5K/month from sponsorships at 5,000+ subscribers plus Pro upsells. This beats a pure SaaS because newsletters build audience ownership and create a natural funnel into the paid product.

3. White-label curation infrastructure for other communities. Target: moderators of subreddits, Discord servers, and niche forums who want their own "HackerTube for X." Revenue: $99–499/month per instance. This beats building more consumer sites because you sell infrastructure to people who already have distribution — and the same scraper and ranking engine serves every customer.

Product Ideas

🥇 HackerIndex — "Search every video, book, and paper Hacker News has ever recommended." Target: developers and technical researchers. Why now: the copycat wave proves demand for curation, but every existing site is stateless and unsearchable. A persistent, multi-format index with an API is the durable version of what Karpathy demoed. Charge $8/month Pro, $99/month API.

🥈 CurateKit — "Spin up your own community-curated discovery site in 10 minutes." Target: subreddit mods, Discord community owners, niche newsletter writers. Why now: HackerBooks and HackerTube clones are being hand-built by individual developers; a white-label tool captures that demand as recurring revenue instead of competing for the same eyeballs. Charge $49–299/month per instance.

🥉 Signal Digest — "The weekly best of HN-recommended content, in your inbox." Target: busy senior engineers who want the signal without the thread. Why now: the digest is the cheapest possible MVP (no UI, no search, just curation and email), validates willingness to pay fastest, and builds the subscriber list that feeds the other two products. Free tier plus $5/month premium with archives.

SEO Opportunity

Search volume is near zero today — nobody searches "HackerTube" yet, and that's the point. The play is long-tail intent queries that will grow as the category does: "best videos recommended by Hacker News," "HN recommended books list," "curated technical video search," "Hacker News video archive," "search HN comments for YouTube links." SEO difficulty is effectively 0/100 — no established sites target these terms. Content strategy: build one indexable page per topic cluster (e.g., "Best system design videos according to HN"), auto-generated from your data with real context snippets. That's hundreds of long-tail pages from one dataset, and each one is genuinely useful rather than thin filler.

Risk Assessment

The thesis breaks in three ways. Platform risk (highest): HN changes its HTML structure, rate-limits scraping, or YC decides to build this itself. Mitigation: use the official Firebase HN API instead of scraping HTML, cache aggressively, and never depend on real-time access. Market risk: the audience is small and cheap; if conversion to paid lands below 1%, the business doesn't clear opportunity cost. Mitigation: validate willingness to pay with a paid waitlist before building the full product. Execution risk: you build a beautiful consumer site and discover nobody returns after week two because the digest is the actual product. Mitigation: ship the email digest first, before any UI.

Cheap validation: post a "would you pay $8/month for this" question in the HN thread itself, collect 50 emails via a one-page landing site, and offer a $5 pre-order for the first month. If you can't get 20 pre-orders in two weeks, walk away. Walk-away criteria: fewer than 100 digest signups from a front-page Show HN, or any sign that HN is actively blocking third-party access.

Action Plan

Today: Register the domain, stand up a one-page landing site promising "search every video HN has ever recommended," and post a comment in the original HackerTube thread asking what people wish it did. Collect emails.

Week 1: Build the scraper against the Firebase HN API, backfill 12 months of threads, and ship a read-only search UI. Launch the weekly digest to your email list. Post to Show HN on day 7 with the index as the hook, not the UI.

Month 1: Add Stripe billing and the Pro tier at $8/month. Ship the API in beta to 5 users for free in exchange for feedback. Goal: 300 digest subscribers, 25 paying Pro users, $200 MRR.

Month 3: Launch CurateKit as a separate landing page targeting community moderators. Goal: 800 subscribers, 100 Pro users, 3 white-label customers, $1,200 MRR. If MRR is below $500 at month 3 with no growth inflection, stop and redeploy the scraper toward a different community's data.

Related Terms

HackerBooks — the immediate sibling product curating HN-recommended books; validates the multi-format index thesis and is a natural acquisition target or partnership. Show HN launch dynamics — the meta-trend of HN launches as distribution events; relevant because your go-to-market depends on riding the same channel. Anti-algorithm curation — the broader consumer shift toward human-filtered discovery (Kagi, Readwise, Are.na); HackerTube is the technical-video instance of a pattern that's compounding across categories.