In March 2023, OpenAI opened API access to its chat models. Five days later, Tony Dinh, an indie developer in Vietnam, shipped TypingMind, a better frontend for using them. It took $22,000 in its first seven days. By November 2024 it had done $1 million in a trailing twelve months, and by October 2025 it was making $130,000 to $160,000 a month, more than half of it recurring revenue from companies. Every one of those numbers comes from Dinh's own newsletter, which is the reason this story is worth studying: the ledger is public, dated, and includes the parts that failed.
The ledger
- March 2023. TypingMind launches "just 5 days after OpenAI announced the API", per his reflection post. First seven days: $22,000, per his first-million post. It sells as a one-time purchase.
- February 2024. $500,000 in total revenue, about a year after launch. He counts 171 product updates shipped in those twelve months. A subscription tier has reached $15,000 in monthly recurring revenue, quietly passing the $14,000 peak of his previous subscription product.
- November 2024. $1 million in the last twelve months, 20 months after launch. One-time purchases are still about half of monthly revenue. He has closed "one 3,000-seat deal and several 1,000-seat deals" and spent "at least $20K+" on SOC 2, HIPAA, GDPR, security audits and legal work. He reports working about four hours a day.
- October 2025. The latest update: roughly $130,000 to $160,000 a month, with recurring revenue from the B2B Team version now over 50% of the total.
Read the ledger backwards and the shape is clear: a launch spike, a long middle where one-time purchases carried the business, and a second act with entirely different customers. Each stage funded the next, and none of it required an investor, an employee or an office.
Why five days was enough to build it
TypingMind is, in Dinh's description, "a static web app with no back end and no database". That architecture did two jobs. It made a five-day build plausible, because there was no server infrastructure to design. And it made the business model survivable, because a product with almost no operating costs can afford to sell for a one-time price. A subscription exists to cover ongoing cost; when there is none, the one-time purchase is not a discount, it is the honest shape of the deal, and early adopters took it in volume.
The five days also had a decade behind them. Dinh had already built Black Magic, a Twitter tool that reached $5,000 a month in 2022 and sold for $500,000, and a subscription product that peaked at $14,000 MRR. He had an audience that knew his work the day he had something new to show it. The speed is real, but it is the speed of a practiced builder with distribution already standing, which is the same pattern behind Pieter Levels' twelve startups and most of the one-person software companies we have covered.
Individuals bought the tool. Companies buy the paperwork
The most useful part of the ledger is the composition shift. The first year's revenue came from individuals paying once. The growth after that came from selling the same product to companies as a Team plan, and the thing that opened that door was not a feature. It was compliance: the $20,000-plus he spent on SOC 2, HIPAA, GDPR and audits is what makes a 1,000-seat or 3,000-seat deal possible, because procurement departments do not buy software, they buy the paperwork around it. By October 2025 that B2B recurring revenue was more than half the business.
For a solo developer, that sequence is worth memorizing. The consumer version proves the product and funds the boring investment; the boring investment converts the same code into contracts two orders of magnitude larger. It is a different route to the same place Damon Chen reached with Testimonial's price ladder: the first buyers are never where the business ends up.
What did not work
Dinh is unusually plain about the failures. On advertising: "In my latest paid ad campaign, I've spent $600 so far, resulting in only 2 conversions." At $300 per conversion, the channel loses money on almost any self-serve price point, and he says so instead of quietly scaling the budget. His summary of what does work: "Most of my sales are from word-of-mouth." That is not a humblebrag; it is a cost structure. A product bought on recommendation has an acquisition cost of zero, which is what lets a one-person company keep most of what it charges.
The four-hours-a-day schedule deserves the same honest reading. It is a late-stage fact, not a method. 171 updates in the first year is roughly one every two days, and that is what the early pace actually looked like: a founder responding to a brand-new market faster than any roadmap could. The short workday is the reward for that year, not a substitute for it.
The window, honestly
The non-transferable part of this story is the calendar. Shipping in five days only mattered because it was the five days after a major platform opened, when demand existed and alternatives did not. That window closed; others keep opening, and they reward the same preparation: an architecture you can ship fast, an audience that already watches you, and the willingness to sell version one for money in week one. Dinh's own accounting of what the whole arc bought him: "With TypingMind making more than $1M a year, I now have an option to stop working entirely." He keeps shipping anyway.
Discussion
Sign in with Google or just a name. No email link, no password to remember.