Quit at $600 MRR, Then $45K/Month With TypingMind: Every Decision in Tony Dinh's Build-Sell-Release Loop

Note: yen conversions in this article are rough estimates at ¥150/USD.

Timeline

Period	Event	Numbers
Until 2021	7th year as an engineer; salary raised to $8,800/month	—
2021	Quit with Black Magic + DevUtils totaling $600 MRR (2 years of savings, 8,000 followers)	$600 MRR
Until 2022	Black Magic (Twitter analytics tool) grows	Peak $14K MRR
March 2023	Launched TypingMind days after the ChatGPT API opened	$1,000 day one → $2,000 the next day → $10,000 cumulative in 10 days
2023	Twitter API paid-tier crisis → decided within 1–2 weeks to sell Black Magic	Sold to Hypefury for $128,000
Later	Also sold screenshot tool Xnapper	$150,000 (at ~$4K MRR)
As of the interview	TypingMind is the mainstay	$45,000/month (≈ ¥6.75M)

What the Business Is

TypingMind is a web app for using ChatGPT and other LLMs through a better UI. Users bring their own API keys, and the app provides folder organization, chat-history search, prompt management, and more. Dinh positions it as “not a wrapper business but a tool like Postman” — because API costs are not passed through, the design carries no exposure to model providers’ price changes.

The Offensive Call — Initial Velocity Decided Everything

TypingMind shipped within days of the ChatGPT API’s release. It started with a $9 one-time beta price, made $1,000 on day one, and a Product Hunt launch added another $14,000. The biggest moat was not feature depth but being first on the ground at the moment of a platform shift. By the time competitors showed up, he was already ahead on features — folder management, multi-model support (OpenAI/Claude/open-source LLMs) — and on mindshare.

The Defensive Call — a $500K Offer and the Death of an API

The Black Magic sale is a textbook case of platform-risk management for indie developers.

A few months earlier: rejected a $500,000 acquisition offer (had also rejected a $40,000/year offer in the past)
Twitter announces paid API tiers: at minimum $42,000/month in API costs becomes all but certain
Decided to sell within 1–2 weeks of the announcement, selling to Hypefury for $128,000

Twitter later introduced a $5,000/month plan, but Dinh reflects that “if I’d stayed, I would have been chained to Twitter.” Selling for $128K right after turning down $500K — the crux is that he judged by future free time, not sunk cost.

Pricing Philosophy and Way of Working

A crisp rule: “software with no ongoing maintenance cost is one-time purchase; services that need servers are subscription” (DevUtils/Xnapper = one-time, TypingMind = subscription)
The team is just one business assistant and one freelance developer. No documentation, no meetings, no deadlines — a 4-hour workday
The goal is not maximizing growth but balancing a calm pace with “enough money to buy a house without looking at the price”

What to Learn From This

“Quitting at $600 MRR” wasn’t reckless — it was calculated. Southeast Asian living costs, 2 years of savings, 7 years of engineering experience (his self-described “unfair advantage”), and an audience of 8,000 followers. It demonstrates the correct framework for the independence decision: whether you can quit is determined by runway and re-employability, not MRR.

Skill at selling shows up as decision speed. Just as Zenn’s catnose sought a partner 9 days after launch, Dinh completed his sale within two weeks of the API crisis announcement. An indie developer’s real asset is speed of response to change, and “waiting to see how things play out” when bad news lands is itself the biggest risk.

Portfolio “sale proceeds” function as downside insurance. Black Magic $128K + Xnapper $150K = roughly $278K (≈ ¥42M) in sale proceeds funded the next offensive, TypingMind. As with Irie’s MENTA + CLOUD PAPER, running operating income and sale proceeds in parallel raises an indie developer’s survival rate.

“Bring your own API key” is the antidote to platform dependence. The near-death experience of API price hikes at Black Magic was distilled into TypingMind’s design of having users bear costs directly. A fine example of failure being engraved into the architecture of the next product.

Conditions for Reproducing It — and the Limits

Easy to reproduce: “ship fastest right after a platform shift” and “choose the billing model by whether there are maintenance costs” are decision rules you can use starting today
The limits: the initial velocity depended on an existing X (Twitter) follower base. And $45K MRR rode a strong AI-boom tailwind — the risk that model providers offer the same features for free never goes away