On March 6, 2023, Tony Dinh posted a link on Twitter. He built it into a chat client: users bring their own OpenAI API key, and the key is stored on the local device. The reason he made it was that he used ChatGPT every day himself, and what annoyed him most was logging in again, waiting for answers to be typed out slowly, and being unable to search old conversations; so he wanted to solve these annoyances for himself first. Five days earlier, OpenAI had just announced the API for ChatGPT.
By March 10, the tool called TypingMind had received $10,000. On March 11, it launched on Product Hunt, and within one day the number went from 10,000 to 22,000. Dinh subsequently stopped disclosing revenue.
This is worth stopping to look at its original form: the 2023 version of the personal product was static, with no backend, no servers, and no database. Users brought their own API key, and the key was stored on the local device. A product that pushed revenue from $10,000 to $22,000 on its Product Hunt launch day had no backend at all at the very beginning. The following description is based on the current privacy policy and refers to the current default mode, not the original text of the 2023 first version: under the default Local/Static Mode, chat history, prompts, and messages (prompts, messages) stay in the browser; Cloud sync and the Team edition are a separate path: only when users actively turn them on does the selected chat content land on TypingMind's own servers.
What it took over were a few very small actions
For example, a person opens ChatGPT every day, first logs in again, then waits for it to type out the answer one character at a time, and wants to look through last week's conversation but cannot search it. None of these three things is big, but they happen every day. These are exactly what Dinh listed in his 2023 retrospective.
TypingMind gathers these actions into one page. Dinh said early users liked it for four reasons: it was faster than ChatGPT at the time; convenience features such as search, defining roles, and a prompt library; users could use their own API key and pay only for usage, with no monthly fee; and they did not have to log in again every day.
The interface illustration in the official documentation shows answers from several models side by side under the same prompt. This is an illustration from the current documentation, not a screenshot of the 2023 first version; it shows an interface that sends the same question to multiple models and displays their answers side by side.
Image: interface illustration from TypingMind's official multi-model documentation[1]; the original image[2] was retrieved on 2026-11-08, uncropped, retaining the English demo interface.
The privacy policy describes the data paths in detail. Under the default Local/Static Mode, the API key, chat history, prompts, and messages (prompts, messages) are stored in the browser's Local Storage and IndexedDB, and TypingMind has no backend server that reads or collects them; switch to another browser or device, and it is a separate, independent data set. The requests themselves go directly from the browser to servers of providers such as OpenAI and Anthropic, with no TypingMind server in between to view or log them. Some providers do not accept direct browser connections, and users can enable the built-in proxy; the proxy only forwards requests and responses as-is and does not view or log them.
Only when users actively turn on Cloud sync, backup, or sharing, or use the Team edition SaaS, will the selected chats, prompts, messages, AI agents, and bookmarks land on TypingMind's own servers; API keys are not uploaded to TypingMind's servers, while the account, contact details, and license data from purchasing a license are collected separately.
A one-time charge is a calculation
What really made this business take shape was how he charged. Dinh did not build a monthly OpenAI API wrapper—his own term—but a static tool sold as a one-time purchase. He wrote the reasoning plainly: the subscription model is highly competitive, and platform risk is high; and with no servers or databases to maintain, that recurring cost is gone, so a one-time charge can hold up.
I prefer to read this as a calculation, not a posture. A one-time purchase means the same user will not pay again next month, and personal-edition revenue does not automatically compound. He put this calculation on the cost side: since there are no servers to maintain, there is no need to chase users for monthly renewals. The shape of the product and the shape of the pricing are the same thing here.
The costs are also concrete. The key is the user's own; how much they use and pay OpenAI is not something TypingMind touches; in default mode, the same conversation does not automatically sync to another device; the self-hosted version runs on the customer's own servers, data is not transmitted to TypingMind or accessed by TypingMind, and security, backups, and management belong to the customer. These are not bugs; they are what necessarily has to be given up when standing in this position.
Two products that didn't work out, and a person who kept tweeting
Before TypingMind, Dinh made EmojiAI and AskCommand, neither of which succeeded, and together they earned only about $100. When he launched those two products, he already had about 50,000 Twitter followers. AskCommand's promotional tweet got about 5000 likes, but the product still did not take off. His own conclusion was: an audience helps, but it cannot guarantee success.
That sentence becomes more interesting when read later. In TypingMind's early days, 99% of sales came from his Twitter reach, and his marketing strategy was to post a tweet every time he added a feature. There was no growth-hacker-style design; he just kept posting.
At the end of that 2023 retrospective, he said he had hired one freelancer to develop with him, and the next step was Custom Deployment for teams, enterprises, and organizations, moving into B2B. By the October 2025 newsletter, the B2B line had become established: he wrote that TypingMind makes about $130,000 to $160,000 per month, recurring revenue from the Team edition is more than half of monthly revenue, and the business no longer depends on one-time purchases of the personal edition.
In the same piece he said most sales came from word of mouth; the most recent paid ad cost $600 and brought only 2 conversions.
The basis needs to be made clear here: this monthly revenue is a blended figure self-reported by the founder, not MRR and not profit; costs, refunds, customer count, and churn were not disclosed, so how much the business net earns remains unclear. The time window of the $600 ad and the definition of “conversion” were also not given, so customer acquisition cost cannot be calculated.
Another path
If the condition is “use your own key to connect to multiple models,” Chatbox is one that cannot be avoided. It likewise lets users connect GPT, Claude, Gemini, Grok, DeepSeek, and others with their own API key, and they can start for free; they can also subscribe to the Chatbox AI service that includes hosted models.
The difference is not in who can connect models, but in how money is charged: TypingMind's personal edition mainly relies on a one-time purchase, while Chatbox can be used for free with one's own key or subscribed to monthly for its hosted service. These are the two products' respective pricing choices, not three mutually exclusive product paths, and not a question of one replacing the other.
TypingMind's most substantial choice was not building the first version in five days, but eliminating the server so the personal edition could charge only once. The product therefore lives lightly, at the cost that the default personal edition does not automatically sync to another device.