[HN Gopher] Roo or Cline? We're building a superset
___________________________________________________________________
Roo or Cline? We're building a superset
Author : boleary-gl
Score : 58 points
Date : 2025-04-10 09:22 UTC (2 days ago)
(HTM) web link (blog.kilocode.ai)
(TXT) w3m dump (blog.kilocode.ai)
| joshuanapoli wrote:
| Aider is working for me, but it's a simple tool. Am I going to
| experience real benefits from switching to Roo, Cline, or Kilo?
| thawab wrote:
| for my daily work i use aider, but i jump into roo and client
| when there is an update, or someone has an interesting
| workflow. it's like taking a test drive.
| Alifatisk wrote:
| Cool project, another entry to aicoderlist.com
| SparkyMcUnicorn wrote:
| You excited me for a minute, but that list isn't anywhere close
| to comprehensive.
| koakuma-chan wrote:
| Claude Code > all
| boleary-gl wrote:
| Is the thing you like best that it's in terminal? Or something
| more?
| koakuma-chan wrote:
| Just the fact that it's not a VSCode fork or extension
| already makes it superior, but there is more: It is general
| purpose, you can ask it to do pretty much whatever you want,
| not limited to coding. It does not have autocompletion, which
| I believe only gets in the way (suggestions interfere with
| your own thinking).
| danielbln wrote:
| I can have cline control and terminal command, it's by no
| means limited to editing code. On the flip side, if you do
| edit codey it can lean in the LSP and various other IDE
| diagnostics to produce better code more efficiently.
|
| 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).
| koakuma-chan wrote:
| Maybe Cline isn't inherently limited to coding, but you
| still have to open VSCode. Whereas I often use Claude
| Code for things that have nothing to do with text
| editing, for example dealing with ffmpeg.
| koakuma-chan wrote:
| I doubt that LSP and IDE diagnostics are better than
| running your compiler (e.g. tsc) and looking at the
| output, which Claude Code does.
| danielbln wrote:
| Well, that might be alright for Typescript, less so for
| Python. Of course, Claude Code can run the linters and
| type checkers manually, the IDE just does it without the
| extra LLM call.
| koakuma-chan wrote:
| If it's alright for typescript then it's also alright for
| python because running tsc is equivalent to running
| whatever python type checker you use, but yes, there may
| be some overhead because Claude Code has to decide to run
| your compiler/type checker.
| jstanley wrote:
| > suggestions interfere with your own thinking
|
| This definitely happens, but it's worth it. When you and
| Cursor are on the same page the autocomplete is a magical
| experience.
| koakuma-chan wrote:
| I think you're referring to that time when you're writing
| boilerplate and your LLM is able to guess what you want
| to write, and it does feel good, yes, but my point is
| that autocomplete can steer you away from writing unique,
| original and potentially better code than what it
| suggests.
| jstanley wrote:
| It's also good when you're making the "same" change in a
| bunch of different places, after you do it by hand a
| couple of times it gets the idea and you can just keep
| pressing Tab.
| nhumrich wrote:
| It's very good in other ways, but I do enjoy that it's a CLI
| tool and not a full fledged editor.
| behnamoh wrote:
| Yes, price-wise.
| DeathArrow wrote:
| I think agentic features of Cursor are better. I am sticking with
| it for now.
| jrvarela56 wrote:
| I pay for Cursor but the comments here made me question the
| incentives. They have to cut spending. The byo api keys don't
| suffer from that.
|
| Maybe we need evals for these tools? There are benchmarks for
| coding abilities (swe bench?), would be cool to set these
| agentic IDEs.
| bluelightning2k wrote:
| I do not respect this project. "We are stealing and adding no
| value" is not a good pitch, borders on anti open source, and just
| fundamentally why would anyone support you or choose you over the
| original(s).
| fortyseven wrote:
| That stance alone will keep me far away from it.
| repsilat wrote:
| Releasing an open source product that improves on alternatives
| doesn't sound like "stealing" or "adding no value" to me.
|
| They're giving away something valuable, same as the people
| whose work they're building on top of.
| bluelightning2k wrote:
| The linked article literally starts by calling themselves an
| anti innovation company and declaring they do not intend to
| improve on it (at least not in innovative ways)
| submeta wrote:
| I love Cursor, with Claude 3.7 and Gemini. But Claude Code is an
| absolut biest and blows them out the water. It is expensive, but
| it acts like a senior developer. It fixes bugs where Cursor (no
| matter which model) fails to do so, no matter which model.
|
| 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.
| rvnx wrote:
| This is because Cursor tries to save cost on LLM calls, so it
| sends it only a very narrow piece of code, and narrower and
| narrower at every update
| shmoogy wrote:
| Okay glad it's not just me - I've started using cline and was
| about to cancel cursor. Max mode spends a ton of money may as
| well use cline.
|
| If I could proxy Claude code through openrouter I'd be more
| open to testing it ...
| rvnx wrote:
| If you are greedy (or cautious about how you spend, because
| literally RooCode can code 0.5 to 1 USD per request), you
| can enable slow-mode in Cursor, and just ride the wave with
| Sonnet-3.7 for free.
|
| (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.
| Arcuru wrote:
| How much money did you give the Roo and Cline developers?
|
| 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.
| rvnx wrote:
| They claim:
|
| > 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.
| newlisp wrote:
| What is Kilocode's business plan if they don't take a cut on
| API access for third parties?
| stuartjohnson12 wrote:
| To use what they learn from the distribution they get from
| throwing a pile of banknotes in the air to become the
| defensibly best tool on the market, at which point they can
| charge above token price.
| rvnx wrote:
| Until in 2 years once we see new MacBook with AI chips
| and/or lot of memory, and that these chips can load whole
| offline models.
| jjmarr wrote:
| Maybe don't use an open-source licence if you don't want people
| to freely use your project's source code?? Especially a
| permissive one that doesn't require copyleft.
| g_delgado14 wrote:
| The same could be said of corporations who aren't required to
| report pollution and thus exploit because it's legal.
| Morality isn't encoded in most agreements.
| 01HNNWZ0MV43FF wrote:
| AGPL encodes quite a bit of morality, and suddenly when
| expectations are written in clear language people stop
| liking it
| jchw wrote:
| The entire premise of copyleft is taking advantage of
| copyright law to be able to represent moral expectations in
| the copyright license. Within the limits of copyright law
| is plenty of room to choose something restrictive enough to
| fit with whatever you want. If you pick a permissive
| license and people use your software in accordance with
| that license, complaining just doesn't make sense. The
| license _you_ have them _explicitly_ allows this. It 's not
| a loophole, it's not a mistake in the license, it's not
| something that the license authors didn't foresee, it is a
| _feature_.
|
| 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.
| theamk wrote:
| but it is. What do you think the difference between MIT,
| GPL, AGPL and BSL is? It's explicitly what you call
| "morality".
|
| It is super easy to release things under AGPL today. If
| people don't, it is an explicit decision.
| behnamoh wrote:
| this kind of mindset prevents many people from open sourcing
| their projects. "why give it away for free when you can
| monetize the closed source version?" I personally have
| donated to several open source projects but I know it's
| nothing compared to the price I've had to pay for closed
| source ones.
| kgwgk wrote:
| "Why give it away for free when I don't want to give it
| away for free" seems a pretty good mindset to prevent
| someone who doesn't want to give things away for free from
| doing so by mistake and regretting it later.
| benwilber0 wrote:
| Yeah. RooCode intentionally uses one of the most permissive
| licenses in existence and even formally declares in the
| README of the project that they are not responsible for the
| usage or output of the software whatsoever.
|
| And yet people will still get upset when the software isn't
| used like the way they want it to be used...
| davidanekstein wrote:
| Unspoken expectations are premeditated resentments
| pkaye wrote:
| The Kilo Code extension seems to be Apache license.
| ipsum2 wrote:
| It's open source: https://github.com/Kilo-Org/kilocode, so they
| are giving back. I haven't looked into what features they
| offer, since I'm happy with Cline for now
| bluelightning2k wrote:
| Brb just applying to YC with a kilo fork which offers $21 in
| free usage and solemnly swears not to innovate even harder
| ach9l wrote:
| yep, this is the way, i guess. as somebody who has taken this
| very same exercise of cloning cline using cline for my own cline,
| a cline that compiles itself, i've also learned to steal* things
| over the years. i've seen your extension, but i was reluctant to
| give it a try just because it looked just like any other clone,
| but i guess i'll do the same thing again. i've started to see the
| value when i decided to fork and declutter again, this time roo
| code. actually i've perfected forking cline and derivatives with
| my own framework. when you know what you're doing, these tools
| don't put you in the flow. vibe coding done right is another
| level of progress. i've got a cs major though, so i'm a bit
| biased, also helps that i've done masters in theoretical
| computing, theoretical linguistics and machine learning, so i've
| always been attracted to these toys and frameworks, not so much
| to javascript or web development however. this whole exercise, or
| should i say automation? now, takes me back to the days i wrote
| compilers. this is just as fun as code that can compile itself in
| the end. same shit all over again.
|
| 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...
| jawon wrote:
| What size gemma are you using? Is the refactoring running
| independently or managed by you?
| ach9l wrote:
| i've been testing all models that fit the mac studio 512 gb
| ever since i got it. previously i was mostly focused on
| getting tool use and chain of thought fine-tuning for coding,
| around the size of llama 3.2 11b. but even some distill r1s
| on llama 3 70b run well on macbooks, although quite slow
| compared to a regular api call to the closed models.
|
| 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.
| omerfarukak wrote:
| I think you all do wrong; you should focus on focused LLMs like
| JetBrains does. I'm pretty sure you haven't checked the Junie
| yet, but it is soo promising.
| jspdown wrote:
| Could you be more specific? How Jetbrain's approach is
| different?
| omerfarukak wrote:
| approach is similar, but it works better; it can run non-stop
| for minutes and really get the task done.
|
| 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.
| koakuma-chan wrote:
| "Join the waitlist"
| sbarre wrote:
| In the 90s at least when people tried to make money by bundling
| and selling existing free open source software, you got a CD-ROM
| out of the deal.
___________________________________________________________________
(page generated 2025-04-12 23:02 UTC)