Fixing His Own Chat Screen: How Tony Dinh Built TypingMind

Starting from a Frustrating Chat Screen

In March 2023, Tony Dinh used ChatGPT often enough to be bothered by its chat screen. He had to log in again every day, answers appeared slowly, and searching past conversations was hard. On March 1 OpenAI released the ChatGPT API, which opened a way for other developers to connect the same models to their own programs. Tony decided to build the chat tool he wanted to use.

He registered the TypingMind domain on March 2 using a name ChatGPT suggested. He worked on it in spare moments for a few days, gave it a full weekend, and published the first version on Twitter on March 6. At first it was a free tool where users entered their own API key, but as the response grew he added payments, and the first sale came the same day he published it.

A Tool He Built Himself Made $22,700 in Its First Week

TypingMind's users registered an access key issued by an AI provider and paid that provider for the model usage. Tony sold access to the software, which included conversation search, saved prompts, and various integrations. Separating the model usage fee from the software price meant that as customers talked more he did not have to cover the model costs himself, and users could manage model spending against their own usage.

The one-time purchase model also fit the product's structure. The early core product was a static web app that ran in the browser, so running a separate account server or conversation database imposed little burden. On March 10, 2023, the early-bird price Tony announced was $19, rising to $29 the next day at launch and later planned for $39. He kept adding features while giving earlier buyers a lower price.

TypingMind launched on Product Hunt on March 11 and received 902 votes, taking the day's top product. The first week's cumulative revenue was about $22,700. He introduced each new feature on Twitter, and update news became promotion that drove the next purchase. The fast sales drew on the audience he had spent two years building: his Twitter following was about 76,000 at the time. Even so, the AI products he had made earlier, EmojiAI and AskCommand, earned only about $100 combined despite being released when he had roughly 50,000 followers. The same audience produced very different buying reactions depending on the product.

171 Updates in the First Year

OpenAI was improving its own product as well. In a first-anniversary retrospective, Tony explained that OpenAI fixed annoyances such as logging in and slow responses a few months later. Meanwhile he widened the feature set and polished the product, shipping 171 updates in the first year. After getting attention first, adding reasons for customers to keep using it became his main job.

A Team Product and Enterprise Customers

In the summer of 2023, TypingMind Custom, which lets a team build its own AI chat environment, began to take hold as a subscription product. Its price then was $79 a month, and in an early August newsletter summarizing July activity he announced reaching $6,000 in monthly recurring revenue. There was almost no active promotion; a link placed below the main TypingMind screen brought customers in.

Around the same time he also won an enterprise customer with a $20,000 contract. Companies wanted a customized product they could run in their own environment or offer to their own customers, and Tony cleaned up external service dependencies so it would install easily on other servers. A "contact us" button he had been reluctant to add appeared on the pricing page. The plan was to talk with customers directly, understand their needs, and then bundle the shared ones into a product people could buy on their own.

Why companies choose this product shows up in a customer case. The software company Atomic Object valued that employees could try several AI models while using one consistent interface, and that the company could manage models and API keys centrally. The person in charge explained that they did not want to lock the whole organization to one AI vendor. Features for branching a conversation or switching models to compare answers to the same question were used in real work.

Splitting One-Time Purchases and Subscriptions

As subscription products grew, billing split by intended use. The personal core app stayed a one-time purchase, while features that need a server, such as syncing and backups across devices, carried a subscription fee. The team product was also billed on a recurring basis. Because the two products shared core features, improving one could be applied to the other.

In February 2024 TypingMind's cumulative revenue reached $500,000, and subscription monthly recurring revenue was $15,000. Tony explained that subscription revenue helped cover the support costs of existing one-time purchase customers. At the same time he was careful not to casually include features in the personal product that carry ongoing costs. For a lifetime license to hold up, the sales promise and the running costs had to match.

Customers also moved between the personal and team products. The growth path Tony described was that a person buys a one-time license, gets familiar with the product, and some of them lead their company to adopt the team product. Personal sales generated their own revenue and also became a path to experience the enterprise product. He kept raising the quality of the personal product to keep that path alive.

What Enterprise Deals Cost to Prepare, and the Reseller Model

As enterprise deals grew, the sales process lengthened. In November 2024 Tony described running video calls where reaching a contract could take months, and closing one 3,000-seat contract and several 1,000-seat contracts. He handled product explanations and technical questions himself, split roles with colleagues, and pushed to hire dedicated sales staff. Calls and contract management entered a routine that had been all development with no meetings.

Selling to companies also required preparation outside the product. By November 2024 Tony said he had spent more than $20,000 in recent months on security audits, compliance preparation, and lawyers drafting and reviewing contracts. The spending was to meet the items companies review when adopting software from a small team. Alongside improving the user interface, helping buyers get internal approval became part of the business.

Selling through resellers was a way to reduce the burden of dealing with every customer directly. Partners could offer TypingMind under their own brand, bundling pricing suited to the customer with consulting and implementation support. Basic questions were handled by the partner and complex technical issues were supported by the TypingMind team. The team building the product and the seller who understood local customers' work each took their part.

After a Million Dollars: Bringing Models Together

In November 2024, twenty months after launch, Tony announced that trailing twelve-month revenue had reached $1,000,000. One-time purchase revenue was about half of total monthly revenue at that point. Enterprise subscriptions added to the early personal license sales, and the revenue streams grew together.

The product also evolved into a workspace for using several models together. In 2025 Tony announced adding parallel conversations that take answers from several models side by side, background processing, a reasoning mode, and a knowledge base that retrieves external material for answers. Users could reduce the burden of moving between separate tools for each model and keep conversations and materials in one place. The convenience of keeping a familiar workspace even when a particular model's performance changes accumulated into the product's value.

In October 2025 the monthly revenue he published was about $130,000 to $160,000. With recurring revenue from the enterprise team product making up more than half, dependence on personal one-time sales also fell. He wrote that TypingMind had brought him to economic independence, and that he had the option to live on investment returns without working further.

Advertising That Stayed Hard to Solve

Advertising did not come easily even after revenue grew. In the same post Tony said he had recently spent $600 on ads but got only two conversions, and that most sales came from word of mouth. Without a formula for growing by raising ad spend, the satisfaction and recommendations of existing users remained the important sales base.

Agent Skills and the Next Task

In June 2026 he added an Agent Skills feature to TypingMind and showed the product on Product Hunt again after three years. Tony believed that for AI to work usefully it needs enough tools and context, and he named access to a computer or an isolated execution environment as the next development task. A product that began by fixing a chat screen was expanding toward connecting AI to material and work tools.

TypingMind's business found its durability in reducing the management burden that grows for users as model providers multiply. Even when customers try several models, they still have to manage conversation history, prompts, work material, and the team's working environment. Tony gathered that repetitive work into one product and widened the sales range from a personal tool to a company's shared workspace.