Roo or Cline? We're building a superset
How “don’t innovate” will make open source AI coding tools win.
At our company we have a strange mantra: “don’t innovate!” Let me explain.
Zero-innovation company
One approach to building software is to build an innovative product. You think of entirely new features, which users might like better than the current ways of doing things. You then build those features into your product, and hopefully, you were right. But often you weren’t right, and the new features don’t work as well as you hoped. In that case, you iterate, refine, and it might eventually work. Or sometimes, you have to throw it out, and think of something completely different.
In the current landscape of AI tools, there is a lot of innovation already! There are a gazillion different tools, all of which try to stand out. This is great! But if we’re going to add value to the space with another tool, we can’t follow the same strategy.
Instead, we’ve decided to not innovate, but to look at the most successful ideas in other products, and build those as quickly as possible. This is not a new idea—in fact, you could say we’ve not even innovating here 😉:
Steve Jobs famously said (quoting Pablo Picasso): “‘good artists copy, great artists steal’—and we at Apple have always been shameless about stealing great ideas.”
Indeed, Apple famously stole the idea of the graphical user interface from the Xerox PARC research group. And subsequently, Microsoft stole it from Apple. Now those are two of the biggest companies in the world!
Kelsey Hightower updated the quip to “Good developers copy. Great developers paste.”
And in business, this is now called (more charitably) a “fast follow” strategy.
To many people, copying and stealing sounds gross or dirty, as if it’s not right. It feels like cheating on an exam by copying answers from your smarter friend. But it starts feeling different when done in the context of open source. Then you’re not doing this for the benefit of your own private company, but to the benefit of the whole community. After all, we’d make it possible—even encourage—for others to steal right back from us! There are many great examples of this:
Linux copied UNIX and Minix, but as open source.
Android copied many iPhone features, but as open source.
Browsers have had fierce battles with lots of copying between them, with open source winning out (Mosaic/Netscape/Firefox, Internet Explorer/Edge, Google Chrome, Safari).
We’re continuing in this tradition, which, we think, will benefit our users as well as the wider community. Instead of thinking “which new ideas should we try”, we ask “why can’t we have all the best features?!”
So what does that mean for our development and roadmap?
A superset of Cline and Roo Code
A month ago, we forked Roo Code, which itself was a fork of Cline—the most popular AI coding agent on the Open Router leaderboard. Forking is of course the ultimate form of flattery: we think Roo Code is great (and Cline too)!
This meant that on day 1, our set of features was identical to Roo Code’s set of features. Cline itself had already diverged, meaning that it had a mostly overlapping, but still slightly different set of features.
In mathematics there is a concept of a “superset”, which means fully containing a smaller set, and more. Kilo Code was already a superset of Roo Code (since we had the exact same features), so we decided to also become a superset of Cline. For this, we ported over the few features on which Cline had diverged—which was pretty manageable because it was originally the same codebase. We also keep merging in code from Roo Code, to make sure we remain a superset of them as well.
This means that today, Kilo Code is a superset of both Cline and Roo Code’s features. And we intend to keep it this way! On top of this, we’ll keep adding features of our own as well.
Think of Kilo Code as Next.js and Roo Code as React - or Kilo Code as Nuxt and Cline as Vue.js. We’re building on top of great software and improving the user experience - and contributing it back to the open source community!
And so if you’re trying to choose between Clin
e and Roo Code - you should try Kilo Code because we plan to be the best of both worlds!
Unique features from Roo Code:
Temperature control and rate-limiting customization
Internationalization support for 14+ languages
Enhanced prompt capabilities (power steering, etc.)
Per-mode tool selection
Unique features from Cline:
MCP Marketplace integration
Zero config to get started - no need for OpenRouter account setup
Notifications with optional sound notifications for task completion
Our own additions:
Free Claude 3.7 Sonnet access included
"Vibe coding" (auto-acceptance) enabled by default
What’s next?
Of course, this is just the beginning. In fact, this was the easy part! Next, we’ll incorporate all the best features from the most popular tools. We want to fold more projects into “our superset”, meaning that if you’re looking for all of the best features available, simply use Kilo Code. This means you don’t have to configure a bunch of different tools to get the best experience. Ultimately, we would like to take on proprietary projects like Cursor and Windsurf, but in a fully open source way. We would very much like you to tell us which features you would like to see us add first—we’ll then work it as quickly as possible.
And since we’re building this as open source, Cline, Roo Code, and everyone else can steal back from us. That would be super cool! This would mean 1) that “team open source” will gain ground on proprietary solutions 2) that within the open source world, the only way we’ll stay ahead is by shipping faster than everyone else, so users keep choosing us. We’re up for that challenge!
The goal isn't to "win" some AI coding assistant war. It's to build something genuinely useful that makes developers more productive without introducing new headaches. We think that's a mission worth pursuing as a speed run.
Conclusion
Of course, the “don’t innovate” mantra is somewhat facetious. In reality, there is a lot of innovation required to integrate many different features into one coherent product, as well as build everything in a way that allows us to move fast.
Now that we are a superset of Cline and Roo Code, and are only adding more features into the fold, we hope you’d give us a shot:
Try Kilo Code for free in VSCode or Cursor.
Leave feedback on GitHub: we are fanatical about crushing bugs.
Join our Discord to tell us what's important to you.
The free tier includes $20 of Claude 3.7 credits!
And what other AI coding tools should we consider adding to our superset? Let us know! With your help, we’ll proudly steal some more—and make it open source.
