GRAM: A Zed fork without all the AI
114 points by patrik
114 points by patrik
(Why is this story tagged with vibecoding? It's literally the opposite; if anything, people who have vibecoding filtered out are probably going to be more interested in this than those who don't.)
Because that tag is so diffuse so as to be meaningless and it is thus applicable to nearly anything.
It was community suggested/edited, which is becoming a problem IMO.
I recently posted something which was not in any way about vibecoding either but it was still community edited this way: https://lobste.rs/s/x3b4ds/making_end_end_encrypted_ai_chat_feel_like
I think it's because people don't understand how tags work here. They look like topic categories; people apply them like that, like they are hashtags. I bet a lot of users don't know that people use tags to avoid seeing content. I think that's a UI failure of lobste.rs more than a problem with "vibecoding" as a tag (and I have problems with that tag, too)
Here's a submission about a project explicitely created without GenAI, and it's still tagged "vibecoding"?
I was neutral to this tag before, but I feel it's getting used in all kinds of ways that's unclear to me.
about a project explicitly created without GenAI,
It's probably a Ship of Theseus situation, but I think it'd be hard to claim that a fork of Zed which includes Zed's nearly 40,000 existing commits (many of which are the result of GenAI) was created without GenAI. Sure the fork itself is explicitly created with the purpose of never using GenAI, but has also stated they plan to take upstream patches selectively (are patches that were the result of GenAI fair game to be taken?).
I did not tag this submission "vibecoding" nor do I necessarily agree with that tagging, but suspect those that believe any use of GenAI in a projects creation or history qualify for the tag. See the recent thread for prek where only 1.4% of the commits were the result of GenAI and it was tagged as "vibecoding" and I suspect Zed's percentage of GenAI commits in history is far, far greater than 1.4%.
I did not tag this submission "vibecoding" nor do I necessarily agree with that tagging, but suspect those that believe any use of GenAI in a projects creation or history qualify for the tag.
I also think that's the case and I don't think it is wrong per se. Maybe having more granular tags would make things clearer. I also don't know exactly where I stand in this regard (as the creator of the fork). In general I tend to avoid using projects that have been vibe-coded to any extent.
I have no idea why it got tagged with "vibecoding". According to the moderator log, it was due to user suggestions...
It's very confusing. I remember the tag getting added to a story I posted about someone being the victim of someone else's LLMs (https://lobste.rs/s/rfcbij/15_years_later_microsoft_morged_my). I was completely new to the site, so I assumed I just didn't understand stuff, but I see it happening again and again and confusing everyone. Surely a tag shouldn't simultaneously mean "x" and "y, but without all the x" and "someone did x to my y"?!
This is potentially interesting. I was really interested in Zed before they shifted focus and became primarily a chat bot app.
Is it a form of the type "track upstream, rip out unwanted features every release" (à la Waterfox vis-a-vis Firefox) or does it aspire to be its own project that will diverge from its parent project (à la Libreoffice vis-a-vis OpenOffice)?
It has already diverged quite a bit and probably won’t track upstream very closely from here on.
Thanks for creating this fork! I was quite reluctant to try Zed (because of it's overall direction), but Gram looks nice.
Do you have an opinion on the devcontainers feature? From what I can tell, this is not available in Gram. What do you think about porting this over from Zed, or adding some similar "development in a container" feature?
Is it a form of the type "track upstream, rip out unwanted features every release" (à la Waterfox vis-a-vis Firefox) or does it aspire to be its own project that will diverge from its parent project (à la Libreoffice vis-a-vis OpenOffice)?
The author posted this answer today. It is a little too light on the details for me, but it at least directly answers your question.
EDIT: I didn't realize the sibling comment @krig, is the author. My mistake!
I enjoyed the presentation on this at the recent Gleam gathering. I'm a big fan of the concept. Yes, we can all just switch off AI features when presented to us (currently), but I think having the option for something with harder separation is a good thing.
One thing people perhaps aren't considering is that LLM usage can be addictive and easy to over-rely on. For those actively trying to avoid or limit their usage, a fork with it removed is ideal.
Since disabling AI is literally a checkbox that's shown to new users, I'm not sure what difference a separate build makes? And it's still reliant on the Zed maintainers in either case.
I go into it a little bit here, but the main reason for the fork was not the AI, it was the terms of use:
https://gram.liten.app/posts/why/
The AI just made me angry enough to actually do it.
I've read the ToS part of https://gram.liten.app/docs/mission/, that seems like a truly unacceptable license. I know I opened Zed once so I must've clicked "agree"; I guess that means, in Zed's eyes, I'm no longer allowed to look at their source code? Wow. Why would you write terms of use like that for an open source project???
isn't that also a GPL violation? one of the terms is that anyone you distribute binaries to must also have access to the source code, and the same right to modify and redistribute it. if it were possible to circumvent that with a EULA, i think google would have done it years ago
If you are the copyright owner, you can distribute it under whatever terms you want. You are bound by the terms of the GPL when dealing with things licensed to you under the GPL. Nothing that you do to your own independently developed/owned source code can be a GPL violation.
Zed's code is mostly owned by Zed Industries, and they require a CLA of external contributors that licenses their contributions to Zed Industries under terms that basically permit anything they'd want to do. Their use of external libraries is restricted to non-copyleft licenses, so even if the GPL could have restricted them if using GPL libraries (which I'm not sure it could, as long as the user agreement only covered their own code), that doesn't apply.
Why would you write terms of use like that for an open source project???
I think that is a really good question! Maybe their intent is not how I interpret it. I am not a lawyer. But I can't read the text of the license and feel OK about accepting it.
That is written not for you but for the investors. And despite it talking about Editor it's mostly about proprietary services integrated into the editor. The cited restrictions section is only one paragraph the rest is about services. Specifically, collaboration features, their models for edit suggestions, and Anthropic models routed through their servers, maybe more.
They specifically call out open source in the license:
2.3. Open Source Software
Zed makes certain versions of the Editor and related software available at the Zed GitHub Repository: https://github.com/zed-industries/zed (the "Repo"). Your use of such software is subject to the open source software licenses declared in the Repo.
So you can take the source, compile it and be happy to not be bound by the EULA.
AFAICT, so far their binaries were fully compiled from their open source repo with no changes whatsoever. But if you want to be sure you can compile it yourself. It's just one git clone and cargo build --release away. You still get all service integrtions, including collaboration and completions/ai if you want them, just need an account with Zed, which bring you back to Zed End User Terms.
So you can take the source, compile it and be happy to not be bound by the EULA.
That's what I did, yes.
And despite it talking about Editor it's mostly about proprietary services integrated into the editor.
It could say that then. So what is stopping them from changing it to "what they really mean"? TOS updates happen all the time and should be pretty simple. Asking users to agree to what's written on paper while verbally telling them it means something entirely different isn't great.
I think AI integration in a code editor is a bad feature.
Is this because it's typically baked in or are you absolutely against AI integration, even as optional 3rd party extensions?
Personally I am set against AI integration in general, I don't use it and I don't think other people should either.
My reasons are many:
I have my own theories as to why they are so addictive, but they are not scientific. Take them with a grain of salt. But I have a complicated past history with addictive substances of my own, and as soon as I started interacting with AI tools I got a very familiar feeling. Even after I stopped using it, I still caught myself wanting to ask a chatbot as soon as I got stuck on a coding problem.
I think this is probably highly individual. Some people can use cocaine recreationally. I am pretty sure that I am not one of those people.
I agree with your reasons, and I'd like to add to those social and epistemological issues. Also, it's hard if not impossible to use LLMs—even the "open" ones—in an ethical way right now considering how they're sourced and what kind of companies produce them.
These are important issues that people have a hard time recognizing, probably in part due to the conveniences afforded to them by LLM tools that just popped up in their favorite software one day telling them to "Ask me anything!".
But should a text editor be concerned with the myriad issues that revolve LLMs? I'm leaning toward no. I don't want an anti-AI text editor any more than I want an AI editor. As a user I want to have the choice for myself and have control over my own experience.
But should a text editor be concerned with the myriad issues that revolve LLMs? I'm leaning toward no. I don't want an anti-AI text editor any more than I want an AI editor. As a user I want to have the choice for myself and have control over my own experience.
I wouldn't try to stop anyone from doing what they want with the project, of course. I do reserve the right to express my disappointment if someone forks it and adds AI integration back in ;)
I can't speak for them, but my own opinion is: I don't like AI integration in text editors, but I like text editors to be extensible. There's nothing a text editor could or should do to limit a third party's ability to create AI integration extensions.
So I don't like third party AI extensions and would rather they didn't exist, but there's many things in this world I don't like yet don't have the power to do anything about. People who want to make and use third party AI extensions will do that and I don't have to care about them, and I don't have to worry that AI integration takes effort away from the text editor or question the ethics of the core developers.
I like the idea of an opinionated zed alternative a lot. Zeds support for multibuffers is what keeps me using their editor (it seems no other editor has them). So having those multibuffers, but more stable and with less bugs sounds pretty damn good.
No longer accepting vibecoded contributions probably means that the code quality will go up which will make it more attractive to contribute too. Maybe I will go in and fix some of the issues that have been bugging me.
I'm maintaining the zed-editor package for nixpkgs / NixOS and this definitely sparks my interest. While you can indeed disable collaboration and AI features as others have mentioned, they also significantly bloat the compile time of Zed. I was planning to abandon the package once I have found a viable alternative, maybe this will be it.
Edit: I ran an actual build of the nix package with Gram vs Zed for comparison. While the impact isn't as big as I hoped, Zed still takes about 17% longer to build than Gram.
I don't see the point in this. The reason why Zed is doing all the AI is that funding a code editor, no matter how good (you might think it is) is hard. Especially if you're making it open source. So from what I can tell the solution was to trick the VC AI hype into funding it. It adds all the AI stuff but it provides an easy toggle to disable it (source: https://zed.dev/blog/disable-ai-features) and without it they would have less resources to work on everything else.
If you don't want the AI, just enable that setting and move on. If your threat model is based around not trusting that zed's disable-ai feature works then idk why you would use their editor when others that don't have AI exist. It's not like we're running out of editors you can use out there.
Somehow, Neovim, Vim and EMacs are able to get whatever funding they need to keep developing the respective editors without adding AI features.
Maybe being a VC funded for-profit company isn't the right model for a text editor. If your description is correct, they made their deal with the devil. And it has a cost. Maybe, by forking the editor and getting it out of the grasp of the VC vultures, we can retain some of the benefit while distancing ourselves from the cost.
I'll point out that while they didn't add AI features per se - if you go on Neovim's site, the sponsors are: a search engine scraper company, "The AI Code Reviewer", and some other AI code reviewer that doesn't call itself AI until you click through. Scroll below the fold and you'll find the fourth sponsor - "W*rp, built for coding with multiple AI agents". 3/4 of the sponsors are AI companies.
I get that funding open source is hard, and I'm not about to shame anybody for extracting money from AI companies (the more money goes to FOSS projects instead of aibros, the better). I try to view that as strictly a business/survival decision, not an endorsement.
On the other part, Vim seems to be shilling AI just for the love of the game :p
I'm also amazed Emacs is somehow the last one to hold out, considering how AI-friendly its community seems to be.
I'm also amazed Emacs is somehow the last one to hold out, considering how AI-friendly its community seems to be.
Even LLMs balk at elisp...
I know about Neovim's sponsors, I don't like it and it's making me seriously uncomfortable. As long as sponsors only contribute money and don't have any editorial control over the project I haven't viewed it as a crisis just yet, but it has made me think about maybe switching back to Vim. I hear Vim 8 was supposed to be pretty good. Though that link regarding the Vim 9.2 announcement is disheartening.
Maybe being a VC funded for-profit company isn't the right model for a text editor.
Well I wasn't saying that their funding solution was correct. I merely stated what their solution was. I'm not really a fan of VCs and plastering AI into everything either.
If you don't want the AI, just enable that setting and move on.
If you're OK with that and don't mind the terms of use for Zed (like I said in another comment, it was the terms that was the dealbreaker for me, not the AI), I think that's a totally reasonable thing to do.
I fund my hobby projects by working on other things as my day job, so I don't need any outside money. I also don't care about keeping up or adding features, as long as it works for my use and doesn't break. If this project ends up going nowhere after this point I would be totally fine with that.
I had to look for your takes on it* since I never saw any EULA when using Zed on Linux (I'm assuming packagers remove it). I also can't recall if I saw one when I installed Zed on Windows since I used winget, I might've just skipped it and didn't notice.
I do think that the EULA is weird, it reads like a proprietary software license even though the code is open for all to see. Personally I prefer Helix so I never cared much about Zed, I've used it a few times here and there when I couldn't be bothered to set up the LSP for helix, mostly.
What is the value or point in commenting that you don't see the point or value in something? When one or more people create a FOSS project it's clear that they see a point in it, and quite likely, so do others. In fact, the site for the project says why the maintainer has done this. You can agree or disagree, but ISTM that taking the time to chime in "I don't see the point of this" is really just saying "I don't want this to exist" -- which does not make sense to me.
They're scratching their itch and not harming anyone, not even the upstream project. You don't have to use it. I don't have to use it. The whole point of open source is that people can exercise the rights provided to do stuff like this for any reason or no reason at all.
What is the value or point in commenting that you don't see the point or value in something?
It's probably same reason as why content-less comments to the tune of "AI bad" consistently appear and score numerous upvotes under anything vaguely related to it.
Like it or not, humans tend to enjoy expressing their viewpoints.
I like where this is coming from. I wish more text editors stayed in their lanes and didn't intertwine with built-in AI integrations that really should be 3rd party extensions. The text editor should be a stable, performant and modular platform that users can build on.
At least, the AI integrations shouldn't become the core focus of the product (ctrl+f "agent") stealing developer time.
I love Neovim in theory, but I can't get into it. Having something as powerful, free and no-nonsense in the GUI code editor space (so, similar UX to Zed or VSCode) would be fantastic.
I want my web browser to be a web browser, not an "AI browser with AI features disabled". Therefore, I don't want to use Firefox.
Similarly, I want my text editor to be a text editor, not an "AI editor with AI features disabled". Therefore, I don't want to use Zed.
Now I'm fairly happy with NeoVim so far, but Zed seemed pretty good back when it was a text editor. There's ample opportunity for a fork to extract the text editor part of Zed and make it something useful again.
Not a good analogy. Firefox is a "browser with AI features that you can disable", it has nothing built around those features, which is why forks can just track upstream and hardcode-disable them with every release. (But what even is AI anymore? Firefox considers the local translations AI which should be opt-out (because it is, LLM-based genAI), but LibreWolf and Waterfox track it.)
Zed is now built around 3rd party LLM integrations and a fork can't track it.
Even more reason to make an independent fork whose goal is to rip out those 3rd party LLM integrations (and everything else AI).
A fork doesn't have to closely track upstream, Zed is already a good editor ignoring the license and AI stuff. Its current form, minus AI stuff, will remain a good editor.
more open forks like this or VSCodium are always good!
It's just funny to me that the author is apparently anti-ai enough to 1) make a fork of Zed removing all the AI features, and 2) says that "AI makes me angry" on their website, yet the Gleam SSG they're using, Blogatto, is vibecoded. 🤷
That's me :)
Zed is also vibe-coded.
Ah yeah. I mean, I knew zed is vibecoded. Obviously as long as you know Blogatto is vibe-coded as well, it makes sense.
I read a few more comments and I saw that your primary reason for the forking was due to the T&C, which makes a lot of sense actually. cheers!
I'm not super happy about Blogatto precisely because of the vibecoding, but I like gleam and it is an SSG built using some gleam packages I made so it was close at hand for me. I put the site together in an evening using it so OK, fine. It's like the choice to fork Zed. Obviously the purist option would have been not to fork it, but given the choices and conditions (I wanted something to hand to students), it seemed like it could work so I tried it and here we are.
I also take pleasure in turning the enemies weapons against him, if that makes sense. Hence the Le Guin quote on the site: "What cannot be mended must be transcended."
I'm not going to try to get the Zed company to stop vibing or change direction. But thanks to the GPL, I can take what they have built and reshape it into something that works for me. I think they chose the GPL as a way to keep other commercial interests from profiting off their work, but it works in my favour.
Zed shipped an AI kill switch back in July 2025. I'm guessing the GRAM author didn't know about it.
{
"disable_ai": true
}
Not to be excessively snarky, but ChrisDenton shipped this comment 12 hours ago, I guess you didn't know about it.
To answer the question, disable_ai is not good enough for me. I want an editor that I can hand to my students as their first code editor, and I don't want the anti-features that are going to get in the way of them learning to code to be enabled by default. I prefer it if they are not there at all. Maybe a little friction is enough to get them to try something by themselves first before asking the AI for help.
The other answer is that AI is just one on a long list of things I dislike about Zed.
...the list goes on.