Sold
Businesses that were sold. Each story gives the buyer and the price when the sources name them.
Sold holds 17 of the 110 stories on Aiolum: 5 run by a solo founder and 12 by a team of 2 to 10. Together they report US$147M in verified outcomes; the largest is Base44, US$80M cash exit, 18 June 2025. 3 headline numbers carry evidence grade A (disclosed or verified), 2 grade B (independent press) and 12 grade C (founder claim). The median story reached its headline number 11 months after launch. The founders are in 9 countries, most often France and USA. The most common main channel is creators and short videos (4 of 17). The tools named most often in their sources are OpenAI API (12), X (10) and Stripe (8).
One block per story.
Darker earth means stronger evidence.
Verified, number by number
Browse by what the product does
Patterns to copy
More filters: main channel, size and time
The revenue trend compares the latest months in the sources with the months before them. Falling means revenue is about 15% or more below the best recent months, or the sources report a fall. The headline number is the yearly figure or the sale value on each card.
17 of 110 stories shown
You describe an app in plain language. Base44 generates a working web app with its own database, user login and hosting, so you never touch a server.
It launched in the month that AI models became good enough to write a whole app. It sold to people who cannot program, it charged from day one, and it was sold to a company that already had millions of customers.
An iPhone app for makeup. The user uploads a selfie without makeup and a photo of a makeup look, for example from TikTok or Instagram. The app's AI applies that look to the user's own face, then gives a step-by-step routine and a list of products to buy. It is free to download and paid for with subscriptions through the App Store.
Its result is a before-and-after picture of the user's own face, which a short video shows in seconds. The founder chose TikTok as the channel before he built the app, then tripled revenue from almost the same installs by moving and testing the paywall with Adapty.
A web app for real estate agents and photographers. The agent uploads a photo of a room, chooses the room type and a furniture style, and gets back the same room with furniture in about 10 to 15 seconds. It can also remove existing furniture. Plans are monthly subscriptions sized by the number of photos, and the top plan includes an API.
In 2023, general image generators that were asked to furnish a room also changed the room: the founder said DALL-E might turn a window into a wall painting or add a door, which misrepresents a property. VSAI's own model finds the floor, the ceiling, the mirrors and the doors first and only then adds furniture. It cost US$12 to US$69 a month in early 2024, while VSAI's comparison page lists other staging services at US$16 to US$29 per photo with 12 to 48 hours of waiting. Zillow, the listing portal where such photos end up, bought it.
You photograph your meal. Cal AI sends the picture to a rented AI model and returns an estimate of the calories, protein, carbohydrates and fat. No search, no portion menus.
It sold speed to people who gave up on manual food logging, it paid 250 influencers fixed monthly fees to reach them, and it had a 30-dollar-a-year price. The rented model made the hard part cost almost nothing to build.
You speak into your phone or browser. Talknotes transcribes the recording with a speech-to-text model and rewrites it in a style you pick: a clean transcript, a short note, a task list, a formal email, a blog post or a video script. It works in more than 50 languages and exports to text, email or a link.
A marketer who had learned to code one year earlier built it in a week, found that paid ads could buy customers for about US$50 each with a yearly plan that paid the ad bill up front, and sold the business for cash eleven months after launch, when the bugs and the support load became more than one person wanted to carry.
You upload a photo. In about five seconds remove.bg returns the same photo with the background cut away, so the person or product in front stands on a transparent layer. A small preview is free; full-resolution results, the API and the Photoshop plugin cost credits.
It took one task that every designer does by hand in Photoshop, made it one click and free, and let Product Hunt, Twitter and TechCrunch spread it. Then it charged professionals for resolution and for an API, stayed profitable with no investors, and sold to Canva, which had already built the tool into its own product a year earlier.
You upload a low-resolution image, often one made by Midjourney. Magnific enlarges it and invents plausible detail: skin pores, fabric, leaves. A slider sets how much the model is allowed to invent.
One founder had posted AI tutorials on X for a year before the launch, so the audience existed before the product. The tool did something that had no name yet, it charged from the first hour with no free trial, and two people ran it with no employees until the largest design platform in Europe bought it.
A phone and web app that records a lecture, or imports audio, video, a PDF, a document or a link, and makes a transcript, organized notes, flashcards, quizzes, a chat with the note and a short AI-narrated podcast of it. A student gets one free note; after that the app needs a subscription, which starts with a free trial.
Two former Loom staff built a premium-priced app for something students do several times a week, going to class, and sold it through short videos by a small group of contracted creators instead of paid ads. It charged from the first day, and revenue paid for growth with about 50% EBITDA margins until Quizlet bought it.
A tool for people who want to grow an audience on X (then Twitter). It started as a search box over a library of viral tweets, then added a scheduler, automatic direct messages, a contact list, and a language model that proposes tweets and threads in the user's own style. A sister product, Taplio, does the same for LinkedIn.
Two founders shipped eleven products in about four months and kept the one that made money. They sold it on the platform it served, gave an audience owner a share of profits instead of a sponsorship fee, and used a language model to make writing faster, not to replace it. Fifteen months after launch, with about US$1 million a year in revenue, they sold to a larger French software company run by a former schoolmate.
A web page where a user picks a use case (blog outline, product description, email, ad text), a tone and a language, types a few words, and gets a short text written by GPT-3. The free plan gives a monthly character limit; paid plans remove it. Rytr sells the output of OpenAI's model with a simpler interface and a much lower price than the other AI writing tools of 2021.
It entered an existing market with a lower price (US$29 a month for unlimited use, later US$9) and a simpler interface, sold the first customers through AppSumo, and put all marketing effort into search engine ranking, an affiliate program and a reward for reviews. It reached about US$1 million a year in revenue in seven months with three people, and a competitor bought it 18 months after launch.
A web tool that writes marketing text. The user describes a product in a few sentences, picks a type of text (Facebook ad, Google ad, headline, value proposition, blog post, or landing page, the single web page that an advertisement links to) and a language, and gets several drafts written by GPT-3, OpenAI's text model. A document editor has a button that continues a blog post, and a page builder produces a landing page with text, design and code. Each generation costs credits, and the plan sets the number of credits per month.
The founder already sold a small template tool for headlines when OpenAI gave GPT-3 to a first group of developers. He asked OpenAI's chief technology officer for access by email in June 2020, received it, and spent five months rebuilding the tool around the model. When that version came out in December 2020 there were few such tools and he had an audience on X from showing his work in public. He says revenue went from US$1,000 to US$20,000 a month by February 2021. He sold in March 2021, when new competitors appeared every week.
A person describes a web application in plain language. Softgen's AI agent writes a Next.js project with a database, user login and payments, shows a live preview, and changes the app when the person gives feedback. The code goes to the user's own GitHub repository.
A technical founder released an early full-stack AI app builder in July 2024, and revenue stayed flat for four months. Then a second person joined who worked only on distribution, and annual recurring revenue grew about 25 times in four months. They sold the product while the growth was visible and kept their company for the next product.
An app for Shopify stores that sends marketing messages on WhatsApp and answers shoppers with an AI assistant trained on the store's catalog: reminders for abandoned carts, campaigns to subscribers, product advice in 96 languages. The store pays a monthly plan plus a charge for the WhatsApp conversations.
After a first product of their own that nobody bought, the founders copied a kind of Shopify app that already earned money, and sold it in France, where few competitors sold it. They found customers only by cold outreach, with tools that cost about US$380 a month. By their account they sold the company seven months after the App Store launch, when sales outside France became hard.
A web app and a Chrome extension that write LinkedIn posts. The member picks an industry, a niche and a type of topic, or gives an idea or a voice note, and the AI writes a draft in the member's style; the member edits it and posts it. By 2026 it also showed the best posts of top creators, found trending topics on Reddit, X, Google, Bing and Perplexity, and planned posts in a calendar. It never posted on its own.
The founder had already built one of the largest audiences on LinkedIn by posting about AI every day, so the product had buyers on the day it launched. It sold one plan at about US$60 a month, as a US$2-a-day ghostwriter against US$2,000 to US$3,000 for a human one, and his employees grew their own accounts with it and sold it for a commission.
A web form that takes a description of a business idea and returns a long report on its market, customers, competitors, risks and launch steps. In 2023 the text came from hundreds of pre-written prompts sent to OpenAI's ChatGPT models. A later owner still sells it in 2026, rebuilt around live web search and cited sources.
The founders first sold AI reports for many jobs, from lesson plans to travel plans, and kept the one that people paid for: the business idea check. It cost about US$185 to start, so nearly all revenue was profit. An Instagram creator's video in June 2023 brought more than US$10,000 of sales in one week. Six months after launch they sold it through Acquire.com.
A web tool that turns one topic into a topical map: hundreds of search keywords, found and grouped by AI models into clusters of related articles, so that a blogger or an SEO worker can plan a whole site in minutes instead of hours. It sold a monthly number of credits to bloggers, freelancers and agencies.
The founder built it in about three days to remove the part of blogging he hated, keyword research and content planning. Then he took partners, SEO people who had the audience he did not have, and they promoted it. Revenue reached US$7,000 to US$8,000 a month about eight months after the launch, and he sold the product after he had stopped working on it.
A web app that writes and designs a one-page website from a short description. The user typed a few words about a business; the AI wrote the headline and the text, made a logo and illustrations, and filled a page template in about a minute. The page could be shared with a link or downloaded as HTML and CSS code.
It launched in May 2023, when interest in AI tools was at its highest and before Framer, Wix and Webflow had their own AI page generators. It charged from the first day and made about US$8,000 in its first month. Six months after the launch, when sales had stopped growing and the founder no longer cared about the market, he sold it on a marketplace for 2.5 times its trailing-twelve-month figure.
The next verified one-person AI business goes here. Same structure, same evidence grades.
Every number on these cards carries an evidence grade. How Aiolum verifies explains the four grades and the five parts of each story.