Caret's First Release and Origin Story

Hello, I'm Luke, the creator of the AI coding tool Caret. We completed our first deployment at 4:50 AM on July 7th (KST), and many people sent their congratulations and encouragement.

Why did I start Caret?

• A new open-source project challenge after the WebXR metaverse platform XRCLOUD.
• Focusing on AI coding tools like Cursor, and seriously exploring my long-held interest in "Artificial Intelligence."
• Despite questions like "AI in the metaverse?", I wanted to experiment with technology that was one step ahead.

I will continue to develop Caret with the community.

​

Childhood Dreams and Career

Since childhood, watching cartoons, I dreamed of a 'computer that could talk to people.' After graduating around 2008, I worked on AI development in NHN's (now Naver) OCR TF team, where I filed several patents.

I truly felt the potential of AI while working at the Software Policy & Research Institute, with the advent of AlphaGo. However, due to the uncertainty and investment burden of startups, I chose the more tangible field of VR first.

​

Encounter with Cursor and a Sense of Crisis

Last year, as my company faced difficulties, I took on SW development alone, and Cursor was a great help.

In early 2025, when AI agent features were added to Cursor and Windsurf, I sensed a major shift was coming. The question "Could AI replace SW developers?" lingered in my mind, and I saw both a crisis and an opportunity.

​

Cursor was great, but…

After that, I searched for related development methodologies but couldn't find a clear answer.

AI is sometimes a wonderful tool, but at other times, it's frustratingly foolish, and above all, the token cost was a significant burden.

The Cursor Pro plan was insufficient for understanding and testing large projects, and Claude Max or Cursor Ultra cost up to $200 per month. This shows that the primary users of AI are SW engineers.

​

Encounter with Open-Source Cline

While struggling, I discovered an open-source project called 'Cline' and began experimenting with system prompt improvements. However, Cline also had its shortcomings. For reference, Cline is one of the top 5 AI coding tools, alongside GitHub Copilot, Cursor, and Windsurf. As of July 2025, it has over 1.8 million installations on the VS Code Marketplace, more than 47,000 GitHub stars, and is in beta service at Samsung Electronics' DX division.

​

And so, Caret was born

This is the background behind the creation of ‘Caret’.

Here are the initial feature improvements:

​

Why build it myself?

I was often asked, "Why not just contribute to Cline?" But as I learned with XRCLOUD, it's not easy for a Korean developer to participate in a US-centric open-source project due to language and time zone differences. Moreover, Caret is designed with Korean developers and B2B business in mind, so it was necessary to fully reflect their needs.

​

The Advent of the Overlay Architecture

So, I eventually forked Cline and started development. It began with improving the replace logic, but I found out Cline was preparing similar improvements. So, rather than completely separating, I designed an overlay structure that maintains code compatibility with Cline while allowing for independent customization.

This structure has made merging and maintenance much easier, and additional improvements can be made separately. This task alone took me nearly two weeks of sleepless nights.

​

Competition is tough, but...

Can Caret be a better tool than Cursor or Windsurf? Honestly, I don't know yet. But I can confidently say it's better than Cline. I believe it has the potential to be a project that can move faster and more freely than Cline.

I developed most of Caret by myself, and from the midpoint, my university classmate Justin joined as CEO.

​

Sovereign AI and Caret

Some might ask, "Cursor reached $100 million in annual revenue in 14 months, and Windsurf was acquired by OpenAI for $3 billion. What can two people do?" That's right. The objective difference in resources is huge. We don't have the luxury of focusing solely on tech development, and a small amount of granted credits is all we have, so we can't just give away free credits like Cursor.

But isn't the pursuit of Sovereign AI about overcoming this 'war of scale'?

​

The Korean Corporate Way

Korean companies have always found their own way under the premise that 'a head-on competition is impossible.' In the process, they have dominated some markets and shown that they can compete in the global market.

Can Caret do the same? Yes, I'm sure of it. I learned through my experience with Believer that getting investment to grow isn't always the best way. It's more important to do what you can, directly.

​

Practical AI for Developers

Because right now, I, and countless other developers, can't find the 'tool we really need.' Cursor is a great tool, but it's also awkward. The anxiety of sending my code, my concerns, to a company server is something every developer has felt. This is even more true for businesses. Cursor still has difficulty with internal asset configuration, and the system remains opaque.

​

The Starting Point of Caret

So, I started. To create the 'tool I want to use the most.' Perhaps to create my own AGI, Alpha. Caret is not about solving everything with one omnipotent hyper AGI. It's a journey towards an AI that can be optimized for each project, for each developer. An AI as a trustworthy colleague.

​

Finally

AI is not yet perfect. But SW developers want AI more than anyone.

And I believe that such a practical AI coding tool is the best example of the true philosophy of Sovereign AI.

​

Having been in a government agency, I find adding the 'K' prefix a bit cringey, but I ask for your interest and support for the K-AI coding tool, Caret. I especially ask for the interest and contributions of Korean developers. We are currently prioritizing the implementation of login and monetization, and we plan to hold a meeting next week to gather board members. Please try it out and suggest any needed issues on our GitHub issues page.

Thank you.