Tyler's TeardownThe most copied app in AI, and the kid who built it won anyway
A bootstrapped support-bot platform past ten million a year, copied a hundred times over, and the one door every last clone ran straight past.
Here is the most copied product in all of AI, and the uncomfortable part is that the person who built it won anyway. A college student wrapped ChatGPT around your own help docs, shipped it in a few days, and turned it into a company doing something close to ten million dollars a year, while a hundred other people copied the exact same weekend and got nowhere.
If you have never used Chatbase, the idea is almost insultingly simple. You point it at your website and your support docs, it reads them, and you get a chatbot that answers your customers in your own words, dropped onto your site with one snippet of code. Plans run from free up to a few hundred a month, and more than ten thousand businesses now pay for it.chatbase.co
The numbers are not me guessing, because Yasser Elsaid has posted them the whole way up. It did two thousand a month in its first month, sixty four thousand a month within six, passed a third of a million a month in early 2025, and crossed eight million a year later that year on its way toward ten, all bootstrapped, no investors, run by a team of around thirty in Toronto.getlatka.com+1
How it got going is the whole trick, and it is not one you can repeat. Elsaid shipped within days of OpenAI opening its API to the public, so for a short window he was one of the only people offering the thing everyone suddenly wanted, and the launch went off on Twitter while the idea was still novel. Being first for six months was the distribution. That window is shut.indiehackers.com
Could you build it today? In a weekend, and that is exactly the problem, because everybody did. There is a pile of them now, SiteGPT, CustomGPT, Dante, Botpress, Tidio, plus open source clones on GitHub literally named chatbase alternative, some scraping a few thousand a month, most nothing at all. The generic build a bot from your docs product is the single most cloned thing in this entire field.github.com+1
And the floor under all of them is falling out, because ChatGPT will now read your documents and answer questions about them for free, inside a tool a few hundred million people already open every day. When the base version of your product ships as a feature of the platform everyone already has, the base version is not a business. That is the wall, and it is the same wall Photo AI ran into.
So how did Chatbase survive the exact swarm that is drowning everyone else? By refusing to stay a wrapper. It stopped being a thing that answers questions and became a thing that does the work, it now issues the refund, changes the subscription, books the appointment, wired straight into Stripe, Shopify, Zendesk and Salesforce, and it bolted phone support on top. That move is the part worth studying, because it also tells you exactly where your door is.chatbase.co+1
Cloning the bot that answers your docs is a dead weekend, because the winner already walked away from that game and the platforms now give it away for free. The real opening is the one vertical Chatbase is too horizontal to ever serve, its whole support workflow, wired into the tools that trade actually runs on.
Start with what is dead, so you do not waste a month on it. The generic answer your questions from your docs bot has no floor left. Fifty clones are giving it away, ChatGPT does it natively, and a customer can stand up a passable one themselves in an afternoon. Undercutting Chatbase with a cheaper version of that is the Photo AI mistake in a new coat, racing to zero in a market the platforms are about to flood. That specific door is a wall. But unlike headshots, there is a real door standing right next to it, and it is the one Chatbase itself showed you.
Look again at how Chatbase pulled away from its own clones. It was not a better answer engine, it was the actions, the moment the bot stopped talking and started doing, issuing the refund, rescheduling the appointment, updating the account, reaching into a company's real systems to change something. That is the part a weekend clone cannot fake, because it is integration work and earned trust, not a prompt. But here is the tell. Chatbase's actions are horizontal, generic connectors to Stripe and Zendesk and Salesforce that work for everyone and therefore fit no one exactly. It has to stay general, because it sells to everyone.
That is the gap, and it is not another horizontal bot, it is the opposite. Pick a single trade where support is high volume, repetitive, and runs on specific software the generic connectors do not speak. Dental offices drowning in booking, recall and insurance questions, tied to the practice system they already run. Home services fielding the same scheduling and dispatch calls all day, sitting on top of their field service tool. Property managers answering the same maintenance requests that belong in AppFolio or Buildium. Build the support agent that does that one world's actual jobs, hooked into that one world's actual tools, and speaks its exact language.
The reason this stays open is the same reason it stayed open in every other teardown here. A thirty person team racing toward enterprise, voice and bigger logos is building breadth, more channels, more generic connectors, more features for the Fortune 500. It is never going to stop and learn the exact escalation rules of an HVAC dispatcher or build the one integration a dental chain needs, because that is a rounding error on its roadmap and its whole business depends on staying horizontal. The generality that lets Chatbase sell to everyone is precisely what stops it from going deep for anyone. The space it structurally will not enter is yours.
Distribution follows from the focus, because a vertical means you sell where that trade already gathers, its forums, its Facebook groups, its software marketplace, instead of bleeding out for the words AI chatbot on a Google page Chatbase and fifty clones already own. Be honest about the risks though. The AI part is now the cheap, commodity part, so the entire value lives in the integrations and the domain, and that is genuine, unglamorous work. When a bot takes real actions, reliability stops being optional. And you are betting a year on one trade, so choose one with volume and money in it. None of that closes the door, it just tells you the business is in the plumbing and the trade, not the model. Hard, because the obvious clone is a grave, but the vertical support agent that actually does one trade's jobs is a real business a solo builder can still walk into.