Roo Code hit 3 million installs. We're shutting it down to go all-in on Roomote.

[-]

ortegaalfredo@reddit

Sad to hear, but you can always fork the repo.

Roo code is great but I always felt the developers were kinda disconnected from users. I reported some bugs and always got the 'won't fix' treatment.

Now they want to replace it with some slack bot, what are they smoking?

[-]

Right? I saw them shilling hard for zai early on (before their models were even decent) and many people called it out but were promptly banned from the sub. Switched to Kilo at that time and then OpenCode after that went to shit. The most tone-deaf guy on that team is hannes.

[-]

rm-rf-rm@reddit

Going to Cline is going backwards. Move forward to pi or opencode

[-]

wil_is_cool@reddit

I mean both pi and opencode aren't really comparable since they are CLI agents not full human in the loop like roo/Cline, they are kind of just different. Even the new version of Kilo which is opencode based sucks for developers wanting tight agent control.

[-]

runner2012@reddit

Sorry I don't understand. What does Roo cline do that opencode or Claude code can't do?

Normally I run Claude code with a local model and that's worked for me

[-]

Sotyka94@reddit

Isn't it open source? I'm waiting for some other groups to fork it and make their own similar version... Just now that I started to set up custom modes and everything in Roo...

[-]

FullstackSensei@reddit (OP)

I wouldn't hold my breath for anyone to pick up development without some serious funding.

[-]

random-trader@reddit

They will shutdown roomote soon as well. Its nothing new anymore. Everyone is moving into that direction.

[-]

fugogugo@reddit

Jesus.. am I the weird one still using VScode + copilot ?

[-]

FullstackSensei@reddit (OP)

I haven't used coplot in a while, but last time I did, it left me with the impression that a 9B model did most things better.

[-]

Feisty_Resolution157@reddit

Copilot has always been terrible. I’ve had it for free for forever now and I never use it. Microsoft has an uncanny ability to take anyone’s model, mix in their sauce, and make it crap.

[-]

Equivalent-Repair488@reddit

Unrelated but I built a specific flow for excel copilot and they took that away for free enterprise licenses like 2 weeks back.

[-]

mikael110@reddit

It's disappointing but not too surprising, it feels like this is how pretty much all corporate backed OSS AI projects ends up these days. There's not much cash to be made from a truly local OSS product like Roo Code, so naturally they "pivot" to some cloud based garbage instead while pretending it's just a "evolution" of their previous project rather than an entirely different thing.

[-]

hannesrudolph@reddit

Im probably going to get flamed for this.

I work at r/RooCode.

We haven't framed Roomote as an evolution of Roo Code. We've been explicit it's a different product for a different use case, built separately, not at all expecting to take our users with us as much as invite them along if they'd like.

We believe the results will show for themselves (when it comes to picking Roomote over other products) but time will tell. It's a bet we're taking and the idea that people have to agree with it or condemn it is simply not the case.

Roo Code stays Apache 2.0, unused balances refunded, we're pointing users at Cline, anyone who wants to fork can.

We never strung anyone along and never took anything from anyone. We refused to enshittify Roo Code. We know we have been a major contributor to the advancement of this space as a whole through our open source contributions and have loved every minute of it!

I personally am going to miss the community more than anything. It has been my honour working with you.

[-]

rm-rf-rm@reddit

major contributor to the advancement of this space as a whole through our open source contributions and have loved every minute of it!

Can you please enlighten me on this? Roo always appeared to me as a Cursor style re-skin of an existing thing to cash grab stupid VC money. Some micro "improvements" over vanilla Cline does not count

[-]

hannesrudolph@reddit

Cursor-style reskin isn't accurate. Roo started because Cline wouldn't merge our auto-approve PR, so we forked. They eventually shipped that same feature themselves after our version had started chipping away at their user base, which was never the goal. Good relationship with them the whole time, and Cline is the migration path we're recommending publicly.

Right from the jump we diverged. Modes, Orchestrator, custom prompts, and a pile of other features landed in Roo, and pretty quickly it was a different product. We stopped pulling from upstream regularly not long after. The auto-approve push we started with is arguably what forged the way for the "live dangerously" pattern now standard in every code agent. Roo was also born from the community. Major contributions, major engagement, all grassroots. No marketing budget, never bought stars, never bought downloads. 3M installs came the hard way, and that doesn't happen to a reskin.

The extension was free the whole time. No paid tier, no gated features, no revenue play. Hard to cash grab something you didn't monetize. Not a very good attempt at a cash grab if you ask me :p

Shutting down a 3M-install product isn't the cash-grab move. Milking it is.

Are you enlightened now?

[-]

TheThoccnessMonster@reddit

Gottem tbh

[-]

FullstackSensei@reddit (OP)

I don't think you should be flamed for sharing your thinking/insight. I also don't think many thought they're the target audience for Roomote. I, and probably many in the community, are disappointed in the way the project is being left. We understand the license, but there were probably other ways to leave the project.

Regarding the results of Roomote, I doubt that will matter a lot. The moment it's accessible to anyone, they will be able to copy anything that it does better than current offerings. IMO, there's not much of a moat any small team can build in this space, at least not if you're not training your own foundational model with some breakthrough research.

I still wish you all the best.

[-]

rm-rf-rm@reddit

Critical posts on their sub was met with arrogance and derision from that doofus matt. I really hope these charlatans get what they deserve

[-]

FullstackSensei@reddit (OP)

TBH, I also don't understand the logic of this pivot. But then again, we're in a bubble and most AI startups don't make sense to me...

[-]

R_Duncan@reddit

Time to re-check Dirac ( https://github.com/dirac-run/dirac/ ) and cline.

[-]

LirGames@reddit

Initially I used Continue, quickly tested Cline, but ended up switching to Roo Codes and never looked back. This is a bit of a problem. What other options are there that integrate well with VSCode and provide well defined default roles/modes (Architect, Code, Ask, etc)? With good indexing or embeddings? Good context summarization when full... Etc?

[-]

edeltoaster@reddit

I kind of had the same route and wonder the same. Kilo Code is not that good, Continue is not as repiable and flexible in my experience. OpenCode and Pi's integration is meh imho.

[-]

wil_is_cool@reddit

Kilo 3 weeks ago was good, but they rewrote it and removed most of the human in the loop features. As it is now you may as well just a CLI vibe coding framework instead.

[-]

edeltoaster@reddit

That's what I liked with Roo. It did his stuff, but it very clearly asked back if there was clarification needed or checkpoints reached. And it even has buttons for that!

[-]

a_beautiful_rhind@reddit

Lol what? So they're just taking it away just like that? Cline makes models stupid.

[-]

FullstackSensei@reddit (OP)

I genuinely don't understand why they're doing this.

They could have transitioned it into a foundation and let the community take over. By archiving it, they're burning so much goodwill not only from the community, but also from anyone looking to use their enterprise cloud offering.

[-]

a_beautiful_rhind@reddit

Why would you buy the cloud offering when it can change into something else on a whim. Well hopefully it gets forked or its back to trying other agentic frameworks.

A lot of them don't manage context very well and assume cloud models.

[-]

rm-rf-rm@reddit

A lot of them don't manage context very well and assume cloud models.

This was my tell - the more crap and obfuscation a tool adds to the system prompt, the more likely they are the "merchants of complexity" type middlemen trying to extract money they havent earned

[-]

FullstackSensei@reddit (OP)

The cloud space is so crowded. Why buy from this startup and not any of the other who knows how many dozens?

[-]

rm-rf-rm@reddit

I genuinely don't understand why they're doing this.

Money.

hey could have transitioned it into a foundation and let the community take over. By archiving it, they're burning so much goodwill not only from the community, but also from anyone looking to use their enterprise cloud offering.

Does not make money

[-]

Ylsid@reddit

Yes, but also $

[-]

Recoil42@reddit

I worked (briefly) at Roo. There are a lot of problems with the company, and one day I might do a tell all, but that day is not today. For now, I'm just going to say the folks at Roo are very nice people and have a hard time with product direction/vision. It isn't malicious, it's just a really really hard thing to get right.

[-]

FullstackSensei@reddit (OP)

I've worked in the software industry for about two decades now. I've yet to see someone active out of malicious intent. I also understand how hard it is to make a business sustainable when the core product is free and open source.

I wish them the best, but doubt this direction will serve them better.

[-]

Genebra_Checklist@reddit

I was wondering why the model that everyone said worked best on my rtx 5060 ti was utterly dumb while using it with Cline

[-]

Awwtifishal@reddit

Soo... cline vs kilo code vs some fork of roo code? What's best?

[-]

rpkarma@reddit

Pi :)

[-]

rm-rf-rm@reddit

seconded. Watch the dev's AI Engineer Europe talk delivered recently. Its different than all the hype cycle crap out there - feels like it was built by a dude with a good head on his shoulders

[-]

droptableadventures@reddit

I had much better success with Kilo, there's a few major timeout bugs that break Roo with local models.

[-]

FullstackSensei@reddit (OP)

Ironically, I was wondering last week what would it take to extract Roo's prompts and core logic into a minimal rust or go based TUI. Guess I'll have to look into this now...

[-]

0xbyt3@reddit

I am using Pi for the past 2 days; somehow it is faster than RooCode in VSCode and OpenCode. Not sure what it is doing in the background.

I am using it for Unity C# project. Running llama-server and Qwen3.5-9B-UD-Q4_K_XL.gguf on RTX 3060.

[-]

FullstackSensei@reddit (OP)

Pi just gives the LLM shell access and let's it run lose. Of course it's faster because there's no electron and no sandboxing.

[-]

Blues520@reddit

I knew this was coming. I started out my agent journey with Roo Code and hanged out in their Discord. It became clear that they were looking for funding and to pivot to a commercial offering. As I used Roo Code more, I kept getting pushed with their cloud offerings while I was looking to run local.

Then there were the breaking changes. I'd update the extension, only for it to break and then have to roll back to a specific version. They use AI extensively in their github so it was one of those move fast and break things approaches, which can work initially but should eventually have a higher quality focus. Anyways I couldn't run it with Devstral one morning and decided to give OpenCode a try and everything just worked. It didn't break randomly after an update and there was no cloud offering being pushed down my throat so I stuck with it.

I wish them the best though and I will fondly remember Roo as my first venture into agentic development.

[-]

x8code@reddit

The roo community guy is a total asshole, who gets butthurt when somebody provides critical feedback. Mega cringe. I gave up with that extension quite a while ago

[-]

Savantskie1@reddit

So it’s not being maintained, so what? That doesn’t mean you just get rid of it. If it’s solid enough to use now, just keep using it until it no longer works with vscode. I mean wtf? Why do people panic so much.

[-]

FullstackSensei@reddit (OP)

Who's panicking?

It's just sad news, especially the part about archiving the project rather than transferring it to some sort of open source foundation.

It doesn't need to stop working with vs code to stop being useful. If you look at the system prompt today, and what it injects there vs what it was a couple of months ago, you'll notice a difference.

The tools need to evolve as LLMs evolve. It won't take long until roo stops being useful because of this lack of evolution.

[-]

Savantskie1@reddit

You do realize that you can go into the code and augment the system prompts right? lol

[-]

FullstackSensei@reddit (OP)

I already do, but it's not just about prompts. You can probably keep it on life support for quite some time with a ton of your own MCPs, but that makes it into it's own project rather than a tool.

[-]

etaoin314@reddit

im not sure I understand, is there part that depends on their servers it that will break it when they stop supporting it or can we just continue to use roo and it just wont get any more updates?

[-]

FullstackSensei@reddit (OP)

They say they'll archive the project.

In the short term, you can continue to use it. But longer term, anything that breaks won't be fixed and the whole plug-in will rot.

[-]

etaoin314@reddit

ok sucks because I really like it, but at least I dont have to find something new today...still sucks

[-]

Comfortable-Rock-498@reddit

I build this over months as a side project https://github.com/dirac-run/dirac

Please give it a try, https://marketplace.visualstudio.com/items?itemName=dirac-run.dirac or `npm install -g dirac-cli` for cli

It's costs on an average less than half, really fast and scored new high on a single terminal bench I ran (https://huggingface.co/datasets/harborframework/terminal-bench-2-leaderboard/discussions/145) which is yet to be merged.

[-]

ismaelgokufox@reddit

Seems to be a extension very similar to Cline right?

[-]

Comfortable-Rock-498@reddit

Indeed, it originated as a Cline fork (as noted on github). Last I checked, over 67000 LOC were added or changed and another 40k lines were deleted (gutted all mcp, enterprise features etc)

[-]

No_Cattle_4552@reddit

Did YOU build it, or is it AI slop code?

[-]

Technical-Earth-3254@reddit

Wow, that looks horrible lmao

[-]

9gxa05s8fa8sh@reddit

they're cooked. all these goofy bubble startups trying to do something that google would like to get acquired. not gonna fly

[-]

synn89@reddit

I'm not all that surprised. A couple months ago inference broke for us with using FireworksAI and it took 2 weeks for them to fix it. New model support also took ages, even if I submitted the PR. PRs would sit weeks and weeks without review. It was very clear that their dev focus was on other projects.

We ended up moving to OpenCode which has been well supported. When a new model came out I could submit the PR and get it approved and pushed live in a couple hours.

Also, I wasn't impressed with their Roomote and cloud services. They didn't have many back AI providers they'd support, OpenAI API wasn't an option and firing up PR review bot would burn 15 minutes of server time for that 1 action. Meanwhile I could do the same LLM PR review with OpenCode and Github actions, burn 1-2 minutes of server time, and bring any LLM provider I wanted.

[-]