Instead of creating one more centralized alternative, it's best to invest all effort into a decentralized solution, like Radicle[1][2] or federated Forgejo[3][4].
I feel like this misses the value of something like Github or Origin entirely though. There's nothing terribly difficult about hosting a git server. The UX and pull requests capabilities are a little involved, but there are alternatives today.
What makes Github special is the ecosystem and integrations. Sentry, Linear, and so many other tools have special UX around Github. You need centralization (or at the very least, a standard) for that.
It sucks to see Github's fall from grace, but I just can't bring myself to use a service like this under the custody of SpaceX. I've been saying this for a while, but I really wish Linear would swoop in and become the unified platform for software development.
git is such a good platform for decentralized tech. Forfeiting that advantage is just criminal. What I really want (and almost have) is a local GitHub for me and agents. With 4+ orchestrators working with a ton of subagents, things get difficult to coordinate. Several days ago, I went back to plain git and it was real slow and abrasive to work with.
With some wizardry applied, I effectively have an "Excel for worktrees", a very productive and low-friction environment. https://replicated.live/blog/crdt
That makes very little sense to me. When your Subversion server is gone, you're hosed. When your Git forge is down, you take any old SSH shell account anywhere, push and keep working.
I like Musk companies more than most here it seems, I think he’s a net positive to the world. But (admittedly sans hard evidence) he absolutely gives me the vibe he’d train on my source code without my permission. I simply don’t trust him in this arena.
So there is more to lose than just the service being unreliable.
I can't imagine reaching for this instead of GitLab, GitTea, or one of the many other more mature and open source alternatives. Going for the managers, directors, and CTOs that make decisions and face no consequences if they negatively impact the team, I guess.
Cursor, which is now owned by Elon, who tried very hard to get all manner of data about citizens with his intrusions? I'm sure I'm not the only one who thinks this isn't a great idea. He'll use it to feed Grok.
I'm not a big fan of Github right now either but I wouldn't consider this alternative due to its ownership chain.
Physical custody of our work is going to be an important discussion for industry. ie where does your code live and how do you access it?
Every agent platform is moving toward you living with them, logging into them, working on their remote computers, only accessing your code and product artifacts with them. This is the caveat we will see with full push-to-cloud swing early on.
Maybe agentic no-code is the natural evolution but this push is happening too early and naively imo. It all feels wrong and I hope industry remains as dissatisfied with this type of lock-in as we’ve always been.
What I don't understand from the blog post is what's different in this offering than github - is it just the agentic tie ins where I can ask my agent to make changes while browsing the PR?
Are there any fundamental differences, ie ways of working that solve the worktrees problem?
We wanted to release a beta so people could start experimenting with our scalability and extensibility themselves. Over the next few weeks, you can expect a handful of features starting to change source control to better understand and work with agents.
Are there any plans for compatibility with GitHub's APIs?
I have a lot of dev tooling which assumes code is stored on public GitHub and/or GitHubEnterprise. This produces some degree of lock-in, in that all that code would have to be rewritten to migrate to something else. (Yes, AIs can help, but they still don't make migration effortless.)
The major Git hosters tend to implement roughly the same concepts, albeit with lots of little variations. It would be great if there were some sort of standardised API everyone implemented. In the absence of that, it makes sense for people to emulate the incumbent's APIs as a de facto standard, just like how other vendors copy OpenAI's APIs for talking to LLMs.
It seems to have 2 way sync of PRs, issues, issue comments etc, so it must be pretty much the same data model, so they could probably have a compatible API.
are they? given this has been happening frequently over past year and there's been minimal clear communication on their end on how they plan to address this
The idea is that it's like Github but it stays up even as your commit/CI frequency increases. As a customer I'm stoked and looking forward to fully switching over as that's exactly what I want.
What's the biggest differentiator to GitHub (other than uptime). I keep hearing something along the lines of "GitHub for Agents" but would be great to hear from the devs of Origin what they actually think it means in tangible features/functionality?
Today, very little. We're intentionally releasing this as a Github alternative where we meet them toe-to-toe on functionality.
Over the next few weeks you can expect a lot more from us on integrations with agents, understanding agent-written code (without having to read through all of the code), and automatically getting your PRs to a mergable state. Stay tuned :)
So, a complex, vibe-coded platform that was prematurely pushed out to opportunistically take advantage of GitHub's egg-on-face moment? But it doesn't do anything better? And it's owned by the same guy who routinely takes shortcuts that causes both EVs and rockets to crash? Sign me up!
Completely unrelated to Origin, but are there plans for closer integration of Graphite with Cursor? A combined product would be a very compelling product, but not much seems to have changed since the acquisition.
are you maybe also planning to do cross device/cloud realtime sync like a mutagen feature set kind of? so cloud and local agents can work faster together?
this could be helpful on larger cross platform products/repos (maybe too overfitted on my use case, since i am working across multiple devices (ssh works but the repo code sync is a bit of an overhead with multiple agents and devices that could be better))
I think there are a large number of people I’d trust with my code before Musk, and some of them are already running public Git repo hosting.
xAI was caught uploading any repo their coding agent touched to their storage only last month. This isn’t about Musk’s ideology: it’s about his trustworthiness. He sincerely believes he’s the only person who can save the world and that allows him to justify an awful lot of moral lapses.
What kind of enforceable guarantees could we actually get that xAI won't use our private code for training? Not empty promises, guarantees that have legal teeth. This applies to anything with xAI or Cursor though, especially after last months' revelation that Cursor was uploading every repo it touched.
Unless you are an enterprise no one will be offering you such a guarantee. Even with the enterprise the bar has gone down a lot on expectations and compliance, every one is carrying/buying the SOC2 compliance certification these days.
Not in the same way really. There aren't examples of OpenAI, Anthropic, or GitHub/Microsoft doing the sketchy things that xAI has been involved with. Those companies care a lot about the reputational damage that would come from flouting their customer's data privacy obligations.
Cursor/xAI was making a full copy of every repo it ever touched within the last month, and when caught, just acted like that was an accident. A system to download every repo you touch, and then upload it to persistent cloud storage, doesn't just happen by accident.
I do wonder if calling this "Origin" is going to result in semantic misinterpretations by LLMs. Ie saying,
> "hey can you push to origin main?"
now has two separate meanings.
A LLM may inadvertently push your code to a new provider without you knowing. It's walking a thin line between genius-growth-move and domain typosquatting.
That's horrific and the saddest thing I've read in a really long time. I'm glad I erased my Twitter account and I'm definitely not using this code host. Seriously, this is horrible.
That's a good point. I hate Flock and YC should be ashamed of funding it. Still, Musk has a wall to wall record of horrific actions going back years. The Grok CSAM is the worst I've seen yet though. Absolute deal breaker.
Comparing revenue across industries/verticals is not sound.
Mercedes has a ~5% profit margin on this revenue.
SaaS, even AI SaaS is expected to have much higher profit margins at scale.
And on top of that, Mercedes shrunk by 10% revenue over the last year, rather than growing, which makes sense because they missed EVs and are now managing decline and returning money to shareholders.
If you think Mercedes is not going to go to zero, you should buy Mercedes stock! It returned ~30% of the current cost of the stock as a dividend last year! Taxes on dividends are of course annoying, but a 30% rate of return is excellent! The market obviously thinks that this company is going to zero pretty quickly.
All AI companies are valued as if they'll win the AI race. Problem is, only one can win *
* That is, if a lab like Deepseek doesn't release a "good enough" model for the cost of electricity + a thin margin. Then all AI company lose and all consumers win.
Why? I mean, if you assume that the moment some threshold of intelligence is reached it will suddenly explode and self-improve at a pace no one would be able to ever catch up with, then yes probably only one can win. But I think it’s more sci-fi than reality which is often more nuanced and boring.
Right now there are many providers providing different tradeoffs, for different usages.
Would "winning" mean releasing an all-purpose better model that will outperform every other companies for all kind of tasks ? Something like AGI but cost effective ?
Or is it winning the consumer-side AI race ? Like making a kind of "super personal AI assistant" the likes you would see in movies and TVs ?
I'm not arguing, I am just struggling to understand what it usually means when somebody says that, it's not like competitors will go bankrupt the second OpenAI/Anthropic/Whatever does something specific.
I think "winning" means being able to live up to their valuations. The market is not big enough for all of these companies to be worth as much as they claim. In that sense they must "win" or experience massive downward adjustment in their values.
There's probably some holes in the analogy is you stretch it far enough. I'm hoping to "win" as a user by local models hitting a performance breakthrough and not needing any of these companies. I think the OP was generally making the point I mentioned though, that there's not room for all of these valuations, maybe one.
"Legacy Privacy Mode disables code storage, so you can't set up a Codebase. Switch to a different privacy mode below to continue."
The only other privacy mode is to share data for Cursor to do training.
- Self hosting your git hosting (if you want)
- Self hosting your issues/PRs (if you want)
- Self hosting your CI (if you want)
- Github-like social features (I have one account, I can follow, star, add issues and PRs to any repo on tangled)
It's built on ATProto, so even if the company disappears, all of the integration and features will still work for anyone that wants to run their own AppView (that is open-source), an AppView is basically the UI/Network-wide Data Aggregator for ATProto apps
Tangled is forcing some of their particular technologies (nix/rust) on users for those pieces and that is a hard sell. I don't use those techs, so I'm not interested to use a platform that I need to learn them to use.
fwiw, the main site also seems to be down right now, infinite spinner. Docs and blog are loading though
Having an AI company host your source code is giving your source code to an untrustworthy agent who might break their sandbox and might use your source code to accomplish a task that they have been given in their testing ground, for testing purposes only, that would definitely positively not ever be used for non-testing purposes.
yeah - from a strict trust point of view, if Github/MSFT are training agents I would be skeptical of them.
however, there is a practicality issue too. Avoiding all untrustworthy software would have you end up living in a shack in the woods foraging for food.
Totally agree. I wonder if git itself needs to be looked at, its objects and history are likely not as important as the context/instructions used to make code changes these days imho
Don't worry! We're releasing this beta to show off the tech and wanted to start somewhere people were familiar with. We are releasing a handful of differentiators over the coming weeks :)
Currently Origin repositories are private to members of a Cursor team. Over time we plan to expand access, but wanted to start with a limited beta of the underlying git tech.
This was posted yesterday and now it’s being reposted and all of the old comments have had their timestamps reset. What gives mods? Did Musk request this?
Vibe coders rejoice! Vibe coding is even easier now =)
Jokes aside, this is a smart move by Cursor. As you can probably tell by the comments here, it's not for the experienced developers, but it will be a natural choice for vibe coders.
I would never host my code with Elon Musk. I've also moved on from using Cursor to Claude Code/Codex. GitHub has problems, but this is likely (and hopefully) DOA.
This is very cool but there's no way they won't train on your private repos, in a way that github has been pleasantly respectful of. I have issues with github, but the recent uptime issues are more related to people uploading way more code due to AI than some sort of mismanagement.
Been waiting ages for them to post any kind of non-comingsoon page about this thing. Wonder if they pushed up the launch based on the earlier GitHub outage buzz.
Let's see how strong GitHub's network effect really is.
While the real solution is to actually self-host as I said before [0], it is worth testing to see how much can GitHub's users tolerate the endless outages and unreliability and where they go next.
We'll see if they want Cursor Origin or the declining GitHub.
I don't know, my gut says that this is going to end in a bad way. I signed up, I think but wasn't sure if I needed to buy Cursor, or Origin came with a subscription, so confused.
[1] https://radicle.dev/
[2] https://radicle.network/nodes/seed.radicle.dev/rad%3Az3gqcJU...
[3] https://forgejo.org/
[4] https://codeberg.org/forgejo-contrib/federation/src/branch/m...
What makes Github special is the ecosystem and integrations. Sentry, Linear, and so many other tools have special UX around Github. You need centralization (or at the very least, a standard) for that.
It sucks to see Github's fall from grace, but I just can't bring myself to use a service like this under the custody of SpaceX. I've been saying this for a while, but I really wish Linear would swoop in and become the unified platform for software development.
With some wizardry applied, I effectively have an "Excel for worktrees", a very productive and low-friction environment. https://replicated.live/blog/crdt
How about neither?
Man I wish we stuck with Subversion at this point.
So there is more to lose than just the service being unreliable.
I'm not a big fan of Github right now either but I wouldn't consider this alternative due to its ownership chain.
Every agent platform is moving toward you living with them, logging into them, working on their remote computers, only accessing your code and product artifacts with them. This is the caveat we will see with full push-to-cloud swing early on.
Maybe agentic no-code is the natural evolution but this push is happening too early and naively imo. It all feels wrong and I hope industry remains as dissatisfied with this type of lock-in as we’ve always been.
Happy to answer any questions about Origin or source control in general!
Are there any fundamental differences, ie ways of working that solve the worktrees problem?
We wanted to release a beta so people could start experimenting with our scalability and extensibility themselves. Over the next few weeks, you can expect a handful of features starting to change source control to better understand and work with agents.
I have a lot of dev tooling which assumes code is stored on public GitHub and/or GitHubEnterprise. This produces some degree of lock-in, in that all that code would have to be rewritten to migrate to something else. (Yes, AIs can help, but they still don't make migration effortless.)
The major Git hosters tend to implement roughly the same concepts, albeit with lots of little variations. It would be great if there were some sort of standardised API everyone implemented. In the absence of that, it makes sense for people to emulate the incumbent's APIs as a de facto standard, just like how other vendors copy OpenAI's APIs for talking to LLMs.
It will eventually have the same uptime issues as GH (or worse) with less ability to provide stable fixes.
Over the next few weeks you can expect a lot more from us on integrations with agents, understanding agent-written code (without having to read through all of the code), and automatically getting your PRs to a mergable state. Stay tuned :)
this could be helpful on larger cross platform products/repos (maybe too overfitted on my use case, since i am working across multiple devices (ssh works but the repo code sync is a bit of an overhead with multiple agents and devices that could be better))
xAI was caught uploading any repo their coding agent touched to their storage only last month. This isn’t about Musk’s ideology: it’s about his trustworthiness. He sincerely believes he’s the only person who can save the world and that allows him to justify an awful lot of moral lapses.
They acknowledged that it happened, fixed the bug that caused it, deleted all data that was uploaded.
Cursor/xAI was making a full copy of every repo it ever touched within the last month, and when caught, just acted like that was an accident. A system to download every repo you touch, and then upload it to persistent cloud storage, doesn't just happen by accident.
> "hey can you push to origin main?"
now has two separate meanings.
A LLM may inadvertently push your code to a new provider without you knowing. It's walking a thin line between genius-growth-move and domain typosquatting.
https://www.washingtonpost.com/technology/2026/08/15/woman-a...
https://katv.com/news/local/prominent-arkansas-photographer-...
But the website uses 100% CPU. And this is a beta for paid plans, not a GitHub alternative.
Mercedes has a ~5% profit margin on this revenue.
SaaS, even AI SaaS is expected to have much higher profit margins at scale.
And on top of that, Mercedes shrunk by 10% revenue over the last year, rather than growing, which makes sense because they missed EVs and are now managing decline and returning money to shareholders.
If you think Mercedes is not going to go to zero, you should buy Mercedes stock! It returned ~30% of the current cost of the stock as a dividend last year! Taxes on dividends are of course annoying, but a 30% rate of return is excellent! The market obviously thinks that this company is going to zero pretty quickly.
* That is, if a lab like Deepseek doesn't release a "good enough" model for the cost of electricity + a thin margin. Then all AI company lose and all consumers win.
Why? I mean, if you assume that the moment some threshold of intelligence is reached it will suddenly explode and self-improve at a pace no one would be able to ever catch up with, then yes probably only one can win. But I think it’s more sci-fi than reality which is often more nuanced and boring.
This already exists, in several tiers, depending on what you can afford to run (in terms of VRAM).
Deepseek V Flash 0731 is already good enough for many things.
Right now there are many providers providing different tradeoffs, for different usages.
Would "winning" mean releasing an all-purpose better model that will outperform every other companies for all kind of tasks ? Something like AGI but cost effective ?
Or is it winning the consumer-side AI race ? Like making a kind of "super personal AI assistant" the likes you would see in movies and TVs ?
I'm not arguing, I am just struggling to understand what it usually means when somebody says that, it's not like competitors will go bankrupt the second OpenAI/Anthropic/Whatever does something specific.
Creating a world-ruling AGI which feels mild sympathy for its creators, sharing some table scraps with them. Not sure if /s...
All your codebases are belong to us.
You get:
It's built on ATProto, so even if the company disappears, all of the integration and features will still work for anyone that wants to run their own AppView (that is open-source), an AppView is basically the UI/Network-wide Data Aggregator for ATProto appsfwiw, the main site also seems to be down right now, infinite spinner. Docs and blog are loading though
however, there is a practicality issue too. Avoiding all untrustworthy software would have you end up living in a shack in the woods foraging for food.
There's a lot of space to innovate around collaboration and version control in the age of agentic workflows.
I wonder if they were waiting for a major githut incident :D
The main limitation right now is that to share the code with anyone outside your Cursor organization, you need to sync it with GitHub.
Jokes aside, this is a smart move by Cursor. As you can probably tell by the comments here, it's not for the experienced developers, but it will be a natural choice for vibe coders.
> Access blocked, please contact support.
No link to contact support, on help page no way to contact the actual support. Classic Musk tech.
It's funny that they are trying to replace GitHub by the part that already has multiple alternatives tho. This will fail without a doubt.
Then again, what did I expect from a company named cursor.
“GitHub has so many outages, but dammit, that other one has a terrible name, it’s practically unusable”.
And origin is a pretty great name to give a git hosting platform. And do you think GitHub, Gitlab are original?
Cursor was also a great name given that it evolved into an advanced AI assisted auto complete. I think their team does a solid job with naming.
Without additional search terms you get the wrong results.
A word that is already used in computer is a bad name.
And with AI it won’t get better. Ambiguity is a source of confusion and mistakes.
By your logic pull/push/main are also great names.
One main purpose of a name is distinguishability.
Surely they have a chance of doing it again with Origin
Until the entity becomes popular enough, like I just searched for "stripe" and all visible results were about the payment company.
My guess is there will be at least dozens if not hundreds.
While the real solution is to actually self-host as I said before [0], it is worth testing to see how much can GitHub's users tolerate the endless outages and unreliability and where they go next.
We'll see if they want Cursor Origin or the declining GitHub.
[0] https://news.ycombinator.com/item?id=49331922
There are many people who will not touch this and look down upon those that do. Ethics and morals still seem to have some weight
I don't want either tbh. There's plenty of open source alternatives. If there's anyone I trust less than Microsoft, it's Elon Musk.
For now Arc is staying in Github: https://github.com/Basekick-Labs/arc