Roo or Cline? We're building a superset
blog.kilocode.ai
blog.kilocode.ai
I tried Roo with OpenRouter, selected DeepSeek R1, as that was supposed to have the largest output token. Asked Roo to create a documentation for a project I worked on. It created a page of markdown with a few mermaid diagrams. Next I asked Claude Code to do the same. With the same prompt. It created several markdown files with half a dozen mermaid diagrams describing the tech stack, the architecture, the data model and so much more. Another level.
If I could proxy Claude code through openrouter I'd be more open to testing it ...
(or create a new free account and enjoy 500 fast requests)
In theory, you could use RooCode with the GitHub Copilot API too, but in practice there are limits that makes it unusable.
From the website, this is a "well funded" startup that is giving away $20 of credits for free to people who use it. They are literally forking 2 OSS projects and turning them into a funded startup, seemingly without giving anything back.
It is unfortunate when people make decisions that have legal ramifications they do not understand, but there's gotta be at least a little personal responsibility here. It's no different for code than anything else. There was a case not long ago where someone who made royalty free music changed their mind and tried to make copyright claims against people who had used their music in their YouTube videos. But can we be fair here? If you released this thing with an explicit royalty free license like that, can you really get mad because you can't get royalties that you clearly didn't expect in the first place?
If you're worried that you may regret releasing your own works under open source licenses, permissive or copyleft, then simply don't do it. The downside is that you can't be a direct part of the open source ecosystem, either by using existing (non-permissive) libraries or by being adopted by other open source software, but the thing is the vast majority of developers are aware of what they're signing up for. If someone makes a million dollars off of your open source software, you can have an expected paycheck of $0.00, and that even goes if it's copyleft. The plus side of this trade-off is that it enables anyone to take advantage of your software, even if they wouldn't normally have the means of licensing a product for their use case, with very low friction, providing the maximum benefit.
Copyright law, though, certainly does let you encode the expectation that "if you want to make money using this you must pay me", so if that's your expectation, don't use licenses that contradict it, especially not just because it is trendy.
It is super easy to release things under AGPL today. If people don't, it is an explicit decision.
And yet people will still get upset when the software isn't used like the way they want it to be used...
> The free tier includes $20 of Claude 3.7 credits!
> Free Claude 3.7 Sonnet access included
This is what really matters, much more than the IDE itself, they are all VS Code forks (and here, just an extension)
Cursor it's somewhat more interesting than others because they have unlimited premium slow requests, and the fact it is free, makes it acceptable that it's not performing well compared to RooCode.
At the end it's very good because the investors of Cursor agrees to pay with their money your usage of Cursor (if you accept a bit of slow down).
But to be fair, RooCode, or any of the forks, it doesn't matter, as long as the investors pay for the credits.
If Claude would do a $200 unlimited RooCode access, then suddenly RooCode would take a large segment of the market I guess.
If this KiloCode wants to differentiate, this is what they could try.
Its whole project understanding is a lot better.
It has great tool usage; I mean CLI.
There was a project I just would like to create as Vibe Coding, but Claude was not able to complete. Junie fixed it with a single prompt.
Maybe we need evals for these tools? There are benchmarks for coding abilities (swe bench?), would be cool to set these agentic IDEs.
They're giving away something valuable, same as the people whose work they're building on top of.
Claude Code is pretty great, but I find Cline better (plus I can use Gemini 2.5 with it and am not limited to Anthropic only).
This definitely happens, but it's worth it. When you and Cursor are on the same page the autocomplete is a magical experience.
so i gave roo code a try, set a few test cases, and proceeded to declutter, refactor, rewrite the whole thing. i’ve never really written long apps in javascript nor typescript for that matter, and man, i just think 3k lines of code in a single file is just bad code, and i’ve been proven right. 3k lines fucks your context really good. you can’t use cline to code cline because it will ruin you financially one way or another. jesus fuckin’ christ the old cline.ts file was like responsible for the whole damn extension, over 3k lines, the kind of code i would write 10 years ago as an intern. anyway, i’ve added (and learned in the process) react.js components to have an interface to easily collect the data for my own loras. honestly if you are looking to integrate large local models into kilo, i’d love to collaborate. my forks mostly provide data analysis for the fine-tuning of my own personal repositories, using years of commit history as training data, even bash history. i’ve benchmarked several tasks. i can basically fork roo code or cline, declutter it, refactor it, with a gemma or qwq running in a mac studio for a few watts. i’ve been logging everything that i do ever since we were granted api access to gpt3 at a lab i coordinated about 5 years ago. so i’ve mastered the filtering of the completions api, reconstruction of streams, all using airflow and python scripts. i added a couple buttons such as the download task you’ve also added, but more along the lines of “send this to the batch in the datacenter so we train a new gemma” filtering good solutions vs not so good, the old thumbs up thumbs down situation, helps a lot, adding a couple of mcp integrations for applying quick loras locally, plus the addition of test driven development, aiming for reinforcement learning based loras. i built myself a very nice toy, or should i say, i bootstrapped a very nice tool that creates itself? anyway, thanks for sharing this.
i think the next major thing that is gonna happen with these tools is that it gets free at home as new chips become cheaper. llama 4 running in mac studios or dgx stations is as fast as you can get today and it is already good enough (if prepared correctly) to build any yc startup codebase from before covid, or even from before chatgpt, in a weekend. it will definitely happen. i’m wrapping fixing llama4 scout, allow me to mention the fact that it has a tendency to fix bugs by commenting code and adding TODOs, fucking great architecture though, just what we needed, i mean for optimal local development. i’ll try to publish results soon enough, optimized for the top mac studio though, haven’t got a dgx yet. i’ll prepare macbook versions too. the world needs more of this, a cline that fixes itself just on battery power...
for mac studios i've found the sweet spot to be the largest gemma, up until llama scout was released, which fits the mac studio best. scout, although faster to generate, takes a while longer to fill in the long context, basically getting the same usability speeds as with the qwq or gemma 27b.
the refactoring is a test driven task that i've programmed to run by itself, think deep research, until it passes the tests or exhausts imposed trial limits. i've wrote it by instructing gemini, r1 and claude. in short, i've made gemini read and document proposals for refactoring, based on the way i code and strict architectural patterns that i find optimal for projects that handle both an engine and some views such as the react.js views that are present in these vscode extensions.
gemini pro gets it really well and has enough context capacity to maintain several different branches of the same codebase with these crazy long files without losing context. once this task is completed, training a smaller model based on the executed actions, (by that i mean all the tool use: diff, insert, replace and most importantly, testing) to perform the refactoring instructions is fairly easy.