TypingMind: a weekend ChatGPT front-end that grew past $1M a year

What it is

TypingMind is a chat interface for large language models. It started as a new UI on top of the ChatGPT API, "free to use with your own OpenAI API key", built because the default ChatGPT UI logged him out daily, typed slowly and had no search (launch story). It later added Cloud Sync and a custom version for teams, both sold as subscriptions ($500K reflections).

Timeline

• 1 Mar 2023: OpenAI announced the ChatGPT API; Dinh had the idea the same day (launch story).
• 2 Mar 2023: domain bought (name suggested by ChatGPT). 6 Mar: MVP released, first sale the same day. 7–10 Mar: revenue went $1K → $2K → $4K → $10K (launch story).
• 11 Mar 2023: Product Hunt launch, #1 Product of the Day with 902 votes; total revenue reached about $22.7K (launch story).
• Feb 2024: $500K total revenue; the subscription part was at $15K MRR ($500K reflections).
• Nov 2024: $1M revenue over the previous 12 months, 20 months after launch (first million).
• Oct 2025: about $130–160K per month, with B2B team revenue above 50% of the monthly total (Oct 2025 update).

Why it worked

• Distribution before product. Dinh had ~76K Twitter followers; he wrote that 99% of early sales came from his Twitter reach and that tweeting each new feature was his "entire marketing strategy" (launch story).
• Being first to the niche. It shipped five days after the API announcement; "when you're first to the niche, your first version (MVP) doesn't have to be so amazing" ($500K reflections).
• A concrete improvement on a hyped product. Dinh's list of why people loved it: faster than ChatGPT at the time, search, characters, a prompt library, pay-per-use via your own key, and no daily re-login (launch story).
• Pricing matched the cost structure. The core app was static, with no backend, database or accounts, so a one-time license cost almost nothing to serve ($500K reflections).
• Recurring revenue layered on top. Cloud sync and the team version needed servers, so those became subscriptions ($500K reflections).

What to learn

1. A "wrapper" can be a business if it fixes specific daily annoyances that the big product ignores.
2. Let the cost of delivery decide the pricing model: one-time where marginal cost is near zero, subscription where you run servers.
3. Shipping in public turns each feature into a marketing post, and the post history doubles as a changelog.

Risks and what didn't work

• His two earlier AI products, EmojiAI and AskCommand, made about $100 combined, even though one promo tweet got ~5,000 likes (launch story). An audience helps but guarantees nothing.
• OpenAI fixed the original annoyances within months, so the product had to keep moving: 171 updates in its first 12 months ($500K reflections).
• Dinh himself wrote he "underestimated the hype around AI" and could not tell exactly what made it work (launch story), so the launch is hard to replicate without an audience.
• Paid ads did not work for him: in one campaign, $600 produced 2 conversions (Oct 2025 update).
• Lifetime licenses limit which features can be added to the static version without an ongoing cost ($500K reflections).