As someone who has recently tried to refactor our app atop of VSCode (treating it like a platform), we got burned by the UI design decisions that are not straightforward to overcome, let alone maintain. The closed-source MS marketplace did not help either towards our OSS goals.
However, I found Theia (https://theia-ide.org) on HN (like a bunch of other cool things; this is one way I justify the time I spend/sink on this site) and find it a much better fit for our OSS goals (foundation owned, open-source marketplace) with full mod-ability while being compatible to VSCode extensions API (in theory). I recommend you look into it for your app.
You can publish your extensions on OpenVSX fyi. A lot of projects have started doing that now. Not all, but a good amount. Glad you found Theia though.
Ah interesting! I'm building https://double.bot (ai assistant vscode extension) and someone asked about VSCodium but I didn't realize there's a open marketplace for that specifically.
It's made by Eclipse Foundation which was created to host projects around Eclipse the IDE. Kind of like project under Apache brand, Apache foundation and Apache the HTTP server.
Thanks for sharing! If Theia is built on top of monaco, I wonder if a form of one-click switch might work. The monaco editor is theoretically part of the vscode repo, while the "workbench" with settings/configs lives one layer up.
I’d love a good summary on why Cursor, Void (and I assume other tools) decided to build new editors instead of just extending the incumbents (VSCode extension, Jetbrains plugins).
Is there a technical limitation of the extension APIs or is it easier to “market” a new editor than an editor extension?
While adoption for individuals and small companies might be easy, a lot of bigger places already have other integrations with existing IDEs and displacing those for yet another IDE will be a hard ask
> VSCode extensions have very limited control over the UI of the editor. Our Command-K and Copilot++ features aren’t possible as extensions. Same for much of what we want to build in the future!
I don't fancy a future of multiple diverging VSCode forks.
I wonder if they could have created a "Super Plugin" interface instead of a Cursor specific fork and try to get it merged to VSCode or VSCodium or at least keep as a new shared platform for deep customisation instead of balkanisation by everyone interested in doing something similar.
VSCodium does not do any development but only build VSCode codebase to be without all Microsoft telemetry and proprietary stuff and use open-vsx marketplace.
> VSCode
I will leave it as an exercise to the reader to think of some reasons why would VSCode team refuse to accept such change.
Personally I prefer more forks and new editors. Leaving everything in Microsoft’s hands limits how quickly core features can be added. Let the best editor win.
The issue is that the forks don't have the resources to focus on the security issues as they come up, and as the fork gets further and further away from the main project it becomes harder to merge the changes from the main project.
> I’d love a good summary on why Cursor, Void (and I assume other tools) decided to build new editors instead of just extending the incumbents (VSCode extension, Jetbrains plugins).
This is a great question. We think extensions are often buggy (e.g. Copilot, where the selection jumps), or unable to integrate very naturally (e.g. Codeium, other open source extensions). For example, a lot of extensions use the quick pick dropdown for the user to enter a prompt, which feels unnatural. Owning the IDE like Cursor is important to get around these issues, and we think this is why Cursor is so popular.
> While adoption for individuals and small companies might be easy, a lot of bigger places already have other integrations with existing IDEs and displacing those for yet another IDE will be a hard ask
You're right about IDE adoption. We want to build on top of an already-popular to help us reach smaller players, and hopefully we can win over larger companies in the long run with privacy and by centralizing nice AI features.
Cody has done exactly what you said, do it through an extension, and is competitive with Cursor. I've been unable to find a reason to use Cursor over Cody given the "fork" drawback. Though I've mostly switched over to Aider.
Someone else mentioned monetization, which may be the real reason.
Good job at creating more competition, but sorry for being a little bit rude, most software forks die out, try to convince me using this instead of Continue + Claude-dev, which are open source extensions for the vanilla VSCode (or VSCodium).
>Microsoft also made its extension marketplace closed-source
I've tried running 3B LLAMA on my old machine and it was sluggish and kinda bad, but Claude 3.5 Sonnet works great. I will try the 70B model, should be good enough.
I didn't have much success with ollama either. You really need a great accelerator for it to be useful enough. Codestral however is working out great so thanks for the tip.
Congrats on getting it this far! The fun is just beginning, but it does sound like you have an idea of what you're getting into. Here are some thoughts:
There are quite a few extensions and AI editors offering the snippet-like 'tab autocomplete', but Cursor are the only ones afaik who have nailed the full file no-look 'tab-tab-tab-tab autofix' workflow, with what I assume must be a combination of monaco/vscode hackery similar to what you're doing, and their own tuned model (i.e. not an off-the-shelf llm). Is this the piece you're going after when you say 'all of Cursor's core features', or are you primarily focused on something else?
On the topic of extensions, as an occasional user of Cursor I noticed that they've always been able to quite happily 'import' extensions you've already downloaded to a local VSCode profile, which makes sense, but they've occasionally struggled to use the same marketplace APIs for searching for extensions. I'm not sure what you're hoping for here, but I'd suggest you settle for that approach as a starting point. Also be conscious that you're dealing with Microsoft, so be mindful of things like their brand guidelines[0], and avoid telling the world that you're 'hacking away' at their APIs. These things may seem insignificant and it's true that they probably won't cause you a problem unless you're successful.
Lastly, assuming that you're hoping to recruit contributors (which I assume is some of the motivation behind this post): you really need to sanitise your Github workflow and your repo CI. Your commit history[1] currently seems to mostly consist of you updating your docs and committing straight to your main branch with no review process or checks, and the only commit I can see that does have any kind of check in place, your initial commit, has a failed monaco check that you seem to have subsequently just disabled. You'll need to tidy this up a _lot_ and establish a testing baseline if you want to successfully scale the project or attract the calibre of contributor you'll need to be successful.
That is the only reason I use Cursor. The autocomplete is way too good. For refactors, it's just magic! I tried using VSCode after using Cursor for a whole month and Copilot is just straight trash in comparison.
I'd like to continue using Cursor but I'm don't want to have to buy Claude subscription twice (Cursor uses claude under the hood), and I prefer claude sonnet over chat-gpt.
for me very useful feature in cursor is 'apply' where it just merges ai generated code with your code and you just only confirm or reject - for this they also use custom ai and this feature doesn't work with bring in your own key and you need subscription.
I think more competition in the "AI Editor" space is good, but I have to ask – why not make it a VSCode extension? I feel it'd be much more accessible.
Are there limitations in the extension APIs that make it hard to implement?
Agreed, it’s not perfect but VS code is a pretty rich ecosystem. Dealing with the drift between a fork and the core product — and missing features/broken extensions that’ll inevitably come with that isn’t really something I want to deal with in my work flow.
Cursor has a pretty polished experience, but I keep coming back to VS Code and using continue.dev, which is a lot rougher around the edges, because the UI works well enough.
Hey, great question - you're right, extensions are more accessible - the thing is, there is no great way to build the native UI we'd want using them. For example, displaying the before-and-after is vital for us, and doesn't feel natural in an extension.
Extensions like Codeium do the most naive thing possible by writing something like <<<<<<< DIFF >>>>>>>> in your file, literally breaking syntax highlighting. Others like Copilot are super buggy.
Right now Void is built on top of an extension, but we're also modifying and hooking into parts of the IDE to make a more native UX.
A lot of folks prefer JetBrains over VSC. For example my top choice for an AI Editor today is Zed.dev which lets you trivially setup to use a variety of key maps including JetBrains.
Do you need to "own" the whole editor to do what you're doing? Just maintaining sync with upstream is a massive rod for your own back, and I don't see what you gain. As others have said, alternative registries exist, and installation instructions of standalone plugins are easier on users than downloading and installing a whole editor.
Well the code is there: https://github.com/voideditor/void/tree/main/extensions/void
But it's super basic right now... Like literally just the bare minimum to make an LLM call in a side window and a view to apply/reject diffs. Pretty much a weekend project at this point.
I wish someone would fork Omnivim2 instead in order to better compete with VS Code. The editor looks like VS Code, but it is natively compiled and supports VS Code plugins is my understanding. Unfortunately the project seems abandoned. Its coded in ReasonML.
I'm sick of dev tools being built around web browsers. We used to have fully native apps in the 90s and 00s that didn't require so much complexity compared to today.
Edit:
Forgot to mention that it also is powered by vim under the hood, but that's a strength in my eyes.
there is also PearAI - "The Open Source AI-Powered Code Editor. A fork of VSCode and Continue." [0]. It's getting very crowded in this space: cursor.sh, continue.dev, double.bot, supermaven, codium.ai, PearAI and now Void.
I ended up actually not trying double.bot - overall I tried only cursor.sh (first with BYOK and now subscriber) and continue.dev (didn't manage to integrate own key from OpenRouter in JetBrains plugin). For me in the beginning one of the selling point is if I can bring my own key (ideally with GUI setup instead of some config file and working with OpenRouter key). Then if something is open source is also a selling point.
I would recommend update double.bot landing page to have all features shown in few gifs on the same page and 'show in action' short video' - continue.dev had actually pretty good landing page. CodeSnap is probably something similar like cursors' 'apply' but double.bot landing page feels more like documentation. Also at this stage probably worth to add additional comparison to cursor same like you have for copilot
We did a smaller launch a few days ago, but we've actually made a lot of changes to our github contributing page, website, logo since then (based on the feedback). Just trying to ship fast, haha.
I'm surprised no one in targeting Jetbrains IDEs as seriously as Cursor has targeted vscode.
I'd imagine Jetbrains users are much more willing to pay, and the non-AI intelligence Webstorm has baked in save me almost as much cognitive load as Cursor does if not more compared to base vscode.
I tried Continue but it wasn't very smooth at all.
Yeah, my first thought is the plugins are going away sooner or later.
I hope they use Jetbrains as a base instead of VS Code (or roll their own). I strongly prefer IdeaVim or over the one in VS Code. It's the main reason why I don't use Cursor.
There is Cody (https://sourcegraph.com/cody). But their "applying code" feature is worse than Cursor. They put import lines on the bottom of the file, for example. Cursor's "applying code" is superb. Too bad it cannot be used with another IDE.
We're targeting android studio - https://firebender.com. I imagine its because Java Swing is actually extremely painful to work with and learn. Most developers interested in AI and the experience overlap with Sun Microsystem's Java Swing is not very large.
Intellij community edition is open source, but not the enterprise version. However, CE is pretty feature rich with a ton of users and if you're targeting indie developers, most are on CE anyways. I'd say if you do Intellij, you have to be in for the long game to build up to enterprise, but its doable.
That's interesting actually, I didn't know they even had a community edition but as I am reading anyway it seem it supports only Android, Swing, and JavaFX. That's quite a limitation no?
you can try https://firebender.com. Its specifically designed for android studio, but still works well for intellij. I use it every day to develop the plugin itself.
Thanks. I'm already using GitHub CoPilot, but will give this a try.
Honestly speaking I'm not a big fan of chat based assistants, which was one of my complaints about IntelliJ's own assistant. Auto-complete feels more natural, although I can see that the suggestions might be more accurate and to the point with the chat based ones.
Plus points if keeping the tool free for home/OSS users and then charge for corporate or on-premise deployments.
One critical question to the users:
Is Void able to understand the source code files from the project as a whole?
What I mean is being able to read the project, and then make suggestions based on the code that was written and propose new functions? For example, I would see it as fantastic to write unit tests where I'd say: "Please write a unit test that is verifying the functions on ABC.class" or "update the unit tests for ABC.class".
I know it is difficult for an LLM to create a valid result on the first try, but at least it does same me a lot of time creating a skeleton with most of the functions and samples already there. Most times just have to fine tune or fix a few parts but can say it amplifies my output speed by at least 10x to have that kind of functionality.
What is your impression? Can it read all source code files on the project?
> All of the code is OOP-based, and they mount DOM nodes the old-school way (which is what React was supposed to solve..)
I don't know about VS Code, but I remember Atom was refactored to use manual DOM updates because the performance penalty of using React wasn't worth it.[1] By the way, isn't OOP by far the most popular paradigm for building desktop UIs? I imagine VS Code is a difficult codebase to work with that has a lot of intricate code (as is usually the case with large software projects), but that's a strange piece of criticism :-)
I have spent a great deal of time wading through the VS Code codebase and it takes OOP to the extreme. There are mile long inheritance chains, everything is a class, and it is a giant bowl of spaghetti. To some extent, I understand why they used that development approach. It doesn't use a UI framework, just DOM APIs, so classes make sense in lieu of components, but it's still bonkers.
> By the way, isn't OOP by far the most popular paradigm for building desktop UIs?
Yes, but I wish it wasn't. My day job is desktop development (with Electron), and I avoid OOP as much as I can and try to use a functional approach. After jumping all over the VS Code codebase to try to understand how some of this stuff works, and seeing how hard it is to navigate, I think heavy OOP is a bad idea.
Always interesting to hear such claims when graphic editors like Penpot, which have much tighter perf requirements than editors, are so fast while using React.
I think a lot of people don't consider how complex text editors are and then there are the complexities of an IDE, or extensions, to be able to interact with it.
This is cool! I'm a regular user of Cursor (I barely write any code now - just prompt and tab tab tab).
The thing that's really keeping me in Cursor (vs decamping for Zed) is NOT the Cmd K or the sidebar chat. It's actual Cursor's Tab autocomplete model. I've tried many other tab completion models and Cursor Tab blows them out of the water. Supermaven's new release is promising, however.
For the use-case you're solving for (issues with code-privacy), I think something like https://codeium.com/ does already allow for on-prem deployments with enterprise support. I'm trying to think of who would be served well by a fork of VSCode vs Continue.dev or something like a codeium VPS deployment.
Curious: what is your use case for cursor, i.e. what is the problem domain or complexity level (on a scale from boilerplate code to core architecture) that you typically work with? Also what plan are you on and is that enough for your needs?
Hi, we're working with an large typescript codebase and I develop fully fledged features with Cursor which involve non-trivial UI wiring up work. Decent amount of state management using React too.
It's definitely not boilerplate level code - but it also also stops well short of architecture level work. Here's a recent prompt of me trying to implement line-by-line diff generation: https://pastebin.com/WRJpNwqc
When I need to do architectural/design work, I do it in multiple passes, not generating any code, just lots detailed text with tons of feedback/back-and-forth with the model.
I'm on the pro plan but it's not enough for my needs - I run through the quota in 10-15 days. Then, I revert to using Sonnet 3.5 via API key.
Big fan of Codeium here. They're the only player who took vim/neovim users seriously from the start, and it's great to see people building extensions like monkoose/neocodeium. Their models or ux may not always be the best available, but their understanding of their target market and their product positioning is peak IMO, particularly the enterprise/on-prem options you mentioned. My money's on them long-term.
Yeah, I have not given enough time and did so yesterday; it suggests new parameters for example, so I only have to begin a "refactoring" and then tab tab... Vim guys are missing out.
My projects aren’t super complicated and just mostly Python/flask/htmx/html templates. I started with cursor during their free trial so I got the 500 sonnet prompts and it was pretty awesome. Then that ended and the remaining “free” models were fairly lacking.
So I went a’hunting, and settled on the Codeium vscode plug-in free plan which while not as good as cursor/sonnet combination but is quite capable and their “free” models are capable enough for what I am working on (believe they even use a llama 3.1 model— I think they have various models, but left it on the “base model” and it’s been pretty good)
Oh, I’m yet to find a good alternative to Cursor’s RAG-powered side chat. It helps me work with huge codebases so much. Tried Continue, but it’s very unstable, and doesn’t work as well. Would prefer a command line solution, vscode plugin is the next choice, having a separate editor is not ideal, but I’m glad there’s some competition.
Aider lets you pair program with LLMs, to edit code in your local git repository. Start a new project or work with an existing git repo. Aider works best with GPT-4o & Claude 3.5 Sonnet and can connect to almost any LLM.
Thanks! I use Aider with Claude 3.5 Sonnet on smaller projects sometimes, and it's really fun and helpful when it can put a whole repo map into the LLM's context.
I haven't tried this myself so I apologize if it ends up being bad, but I've seen Aider [0] get linked a few times from people who wanted a CLI solution for AI code completion.
Aider is really cool for small projects, but it builds a repo map instead of using RAG. That works on small codebases, but totally fails to be useful on large ones.
>5k source files. They don't fit into the context. I know I can limit what is sent, and I can attach files in the Aider chat myself. But this is not perfect for making an LLM answer questions about a codebase when I don't know much context beforehand. With Cursor, I can just do "@codebase How is a %feature% implemented?", and it's very quick and often helpful with a couple of follow-ups.
Thanks for the suggestion! It needs some work to set up, and it looks like it only works on Github repos. Also, to work with non-local LLMs, you can only use Pinecone for vector storage. I might have misunderstood something, but I will check it out again later.
Don’t listen to the naysayers: this can be Dropbox all over again!
I love Cursor but, it’s a pity it’s not OSS.
The success of Void is tied to how well you implement the part in which AI reasons about multiple files.
This is where all the sauce is in Cursor at the moment imo.
Hey congrats on the release. I don't have much to add other than to ask for more context about this statement-
> allowing AI to edit multiple files at once
Cursor already modifies multiple files at the same time. I typically do this by pressing cmd+shift+I and in the left side of the window the required files will be shown as they're being modified.
I want to build my own agents so I can my private domain specific awareness in the coding environment: PRDs, product docs, API docs, business goals, etc. I'm assuming I could point this to any API compatible endpoint and build my own agents?
You mention refactoring. Do you hope to be able to merge new VScode improvements as they arise, or is this a hard fork that will diverge from VSCode from now on?
I'm not sure why SourceGraph is struggling so much that I have to do it for them (maybe the name?), but Cody is competition and has the benefit of being an extension instead of a fork.
It’s just been so long since going from ‘head to code’ has been the hard part for me. I’d imagine you have to be very unfamiliar for this to be useful.
My primary usage of cursor is cmd+L and cmd+K, I'll give it a shot. Also, makes sense to make an open source alternative. How do you plan on monetizing?
We plan on selling to enterprise companies. We know a lot of companies want to keep their repos private and so we offer them on-prem hosting options. Closed source tools can't really do this because they have to keep their prompt building algorithms behind a private API that they own.
This seems to work pretty well for Codeium, which offers its product free to consumers (at a loss), and is still able to make net profit from enterprise companies. At least that's the plan for now!
I do understand your frustration but again to be pedantic, the type of software isn't conditional to how it's licensing will work.
I think there is a good discussion to be had over the licensing of LLM's and the fruits of training data, as well as that of the training data itself, but when it comes to the software itself, it's hard to argue what is and isn't open source.
> open source would mean that you can also download all the training data and re-do the training yourself
I don't think so.. Utilizing another data source, does not mean they're required, or even permitted in a lot of cases, to distribute that with their software/service.
Sorry but you can pry IDEA from my cold dead hands or at least VSCode doesn't hold a candle to it IMHO. And no, I'm not looking for a discussion on VSCode vs IDEA.
Building a new code editor for a single feature is a dead end in my opinion.
This investment by YC in both Void and Continue.dev perfectly exemplifies the VC spray-and-pray approach by backing two nearly identical open-source VSCode plugins for AI-assisted coding.
They're simply throwing money at similar projects in the same space, hoping one will stick, without any special insight or reason why one should be more successful over another.
Not a bad thing, but just a reflection that among some of these products there is nothing special when compared to rest of the bunch, there are very similar products (even backed by the same investor), and us users have no clear idea what is different between them, if anything, and why we should use one over another.
I guess similar theme can be seen in LLM/AI space in general, or previously in the streaming services, just throw as much money at the problem in the hope that your solution will stick, without any significant and clear differentiation.
Just the two that I have recently been checking out and, as many other commenters in this thread asked, what is the difference between the two? I guess investors know something we don't, or they are just throwing money at the wall...
What might they be? API calls, read all files, write to files, side bar, UI interactions in editor. These are things I would imagine an extension can do. Continue.dev is an example.
As someone who has recently tried to refactor our app atop of VSCode (treating it like a platform), we got burned by the UI design decisions that are not straightforward to overcome, let alone maintain. The closed-source MS marketplace did not help either towards our OSS goals.
However, I found Theia (https://theia-ide.org) on HN (like a bunch of other cool things; this is one way I justify the time I spend/sink on this site) and find it a much better fit for our OSS goals (foundation owned, open-source marketplace) with full mod-ability while being compatible to VSCode extensions API (in theory). I recommend you look into it for your app.
You can publish your extensions on OpenVSX fyi. A lot of projects have started doing that now. Not all, but a good amount. Glad you found Theia though.
Ah interesting! I'm building https://double.bot (ai assistant vscode extension) and someone asked about VSCodium but I didn't realize there's a open marketplace for that specifically.
Is Theia related to Eclipse (the traditional Java IDE)? Or is the reference to "eclipse" just an unfortunate naming collision?
It's made by Eclipse Foundation which was created to host projects around Eclipse the IDE. Kind of like project under Apache brand, Apache foundation and Apache the HTTP server.
They created it, but the code base is unrelated.
Theia is the Monaco editor from VS Code with a different ui
Thanks for sharing! If Theia is built on top of monaco, I wonder if a form of one-click switch might work. The monaco editor is theoretically part of the vscode repo, while the "workbench" with settings/configs lives one layer up.
I’d love a good summary on why Cursor, Void (and I assume other tools) decided to build new editors instead of just extending the incumbents (VSCode extension, Jetbrains plugins).
Is there a technical limitation of the extension APIs or is it easier to “market” a new editor than an editor extension?
While adoption for individuals and small companies might be easy, a lot of bigger places already have other integrations with existing IDEs and displacing those for yet another IDE will be a hard ask
Extracting from cursor team [1]
> VSCode extensions have very limited control over the UI of the editor. Our Command-K and Copilot++ features aren’t possible as extensions. Same for much of what we want to build in the future!
[1] https://forum.cursor.com/t/why-not-a-plugin/2448/2
I don't fancy a future of multiple diverging VSCode forks.
I wonder if they could have created a "Super Plugin" interface instead of a Cursor specific fork and try to get it merged to VSCode or VSCodium or at least keep as a new shared platform for deep customisation instead of balkanisation by everyone interested in doing something similar.
> VSCodium
VSCodium does not do any development but only build VSCode codebase to be without all Microsoft telemetry and proprietary stuff and use open-vsx marketplace.
> VSCode
I will leave it as an exercise to the reader to think of some reasons why would VSCode team refuse to accept such change.
Personally I prefer more forks and new editors. Leaving everything in Microsoft’s hands limits how quickly core features can be added. Let the best editor win.
The issue is that the forks don't have the resources to focus on the security issues as they come up, and as the fork gets further and further away from the main project it becomes harder to merge the changes from the main project.
This is just Eclipse Theia plugins. In fact, I wouldn’t be surprised if this project is actually Eclipse Theia based
> I’d love a good summary on why Cursor, Void (and I assume other tools) decided to build new editors instead of just extending the incumbents (VSCode extension, Jetbrains plugins).
This is a great question. We think extensions are often buggy (e.g. Copilot, where the selection jumps), or unable to integrate very naturally (e.g. Codeium, other open source extensions). For example, a lot of extensions use the quick pick dropdown for the user to enter a prompt, which feels unnatural. Owning the IDE like Cursor is important to get around these issues, and we think this is why Cursor is so popular.
> While adoption for individuals and small companies might be easy, a lot of bigger places already have other integrations with existing IDEs and displacing those for yet another IDE will be a hard ask
You're right about IDE adoption. We want to build on top of an already-popular to help us reach smaller players, and hopefully we can win over larger companies in the long run with privacy and by centralizing nice AI features.
Cody has done exactly what you said, do it through an extension, and is competitive with Cursor. I've been unable to find a reason to use Cursor over Cody given the "fork" drawback. Though I've mostly switched over to Aider.
Someone else mentioned monetization, which may be the real reason.
Good job at creating more competition, but sorry for being a little bit rude, most software forks die out, try to convince me using this instead of Continue + Claude-dev, which are open source extensions for the vanilla VSCode (or VSCodium).
>Microsoft also made its extension marketplace closed-source
Ehm, Open VSX Registry exists.
Are you using a local model with Continue?
I've tried running 3B LLAMA on my old machine and it was sluggish and kinda bad, but Claude 3.5 Sonnet works great. I will try the 70B model, should be good enough.
I didn't have much success with ollama either. You really need a great accelerator for it to be useful enough. Codestral however is working out great so thanks for the tip.
Yeah, I’ve found local models are completely useless for coding tasks whereas openai and Claude can at least do some refactors with guidance
Congrats on getting it this far! The fun is just beginning, but it does sound like you have an idea of what you're getting into. Here are some thoughts:
There are quite a few extensions and AI editors offering the snippet-like 'tab autocomplete', but Cursor are the only ones afaik who have nailed the full file no-look 'tab-tab-tab-tab autofix' workflow, with what I assume must be a combination of monaco/vscode hackery similar to what you're doing, and their own tuned model (i.e. not an off-the-shelf llm). Is this the piece you're going after when you say 'all of Cursor's core features', or are you primarily focused on something else?
On the topic of extensions, as an occasional user of Cursor I noticed that they've always been able to quite happily 'import' extensions you've already downloaded to a local VSCode profile, which makes sense, but they've occasionally struggled to use the same marketplace APIs for searching for extensions. I'm not sure what you're hoping for here, but I'd suggest you settle for that approach as a starting point. Also be conscious that you're dealing with Microsoft, so be mindful of things like their brand guidelines[0], and avoid telling the world that you're 'hacking away' at their APIs. These things may seem insignificant and it's true that they probably won't cause you a problem unless you're successful.
Lastly, assuming that you're hoping to recruit contributors (which I assume is some of the motivation behind this post): you really need to sanitise your Github workflow and your repo CI. Your commit history[1] currently seems to mostly consist of you updating your docs and committing straight to your main branch with no review process or checks, and the only commit I can see that does have any kind of check in place, your initial commit, has a failed monaco check that you seem to have subsequently just disabled. You'll need to tidy this up a _lot_ and establish a testing baseline if you want to successfully scale the project or attract the calibre of contributor you'll need to be successful.
Hope that helps, best of luck with the project!
[0]: https://code.visualstudio.com/brand [1]: https://github.com/voideditor/void/commits/main/
That is the only reason I use Cursor. The autocomplete is way too good. For refactors, it's just magic! I tried using VSCode after using Cursor for a whole month and Copilot is just straight trash in comparison.
I'd like to continue using Cursor but I'm don't want to have to buy Claude subscription twice (Cursor uses claude under the hood), and I prefer claude sonnet over chat-gpt.
for me very useful feature in cursor is 'apply' where it just merges ai generated code with your code and you just only confirm or reject - for this they also use custom ai and this feature doesn't work with bring in your own key and you need subscription.
I think more competition in the "AI Editor" space is good, but I have to ask – why not make it a VSCode extension? I feel it'd be much more accessible.
Are there limitations in the extension APIs that make it hard to implement?
Because it's difficult to monetize extensions and justify a large VC valuation...
(Void is HN backed)
Are you HN backed when your Show HN gets over 200 votes?
The linked website says "Backed by Y Combinator" lol
Agreed, it’s not perfect but VS code is a pretty rich ecosystem. Dealing with the drift between a fork and the core product — and missing features/broken extensions that’ll inevitably come with that isn’t really something I want to deal with in my work flow.
Cursor has a pretty polished experience, but I keep coming back to VS Code and using continue.dev, which is a lot rougher around the edges, because the UI works well enough.
Hey, great question - you're right, extensions are more accessible - the thing is, there is no great way to build the native UI we'd want using them. For example, displaying the before-and-after is vital for us, and doesn't feel natural in an extension.
Extensions like Codeium do the most naive thing possible by writing something like <<<<<<< DIFF >>>>>>>> in your file, literally breaking syntax highlighting. Others like Copilot are super buggy.
Right now Void is built on top of an extension, but we're also modifying and hooking into parts of the IDE to make a more native UX.
A lot of folks prefer JetBrains over VSC. For example my top choice for an AI Editor today is Zed.dev which lets you trivially setup to use a variety of key maps including JetBrains.
https://zed.dev/
Zed looks promising, I'm waiting on editorconfig support PR to drop
Zed is very fast but has a long way to go still.
ssh server is still in early stages, not even p2p yet.
The requirement for extensions to be built in rust only makes it harder to adopt.
When they these 2 things get solved and the editor remains fast, it might actually be very tempting to replace vscode with zed.
Vscode is fantastic but tends to become bloated quickly....
The tradeoffs are certainly there. I'm expecting it to be a viable switch in a year or two.
Idk, continue is absolute trash at the moment. It manages to break my entire VScode environment, including vim hotkeys.
Do you need to "own" the whole editor to do what you're doing? Just maintaining sync with upstream is a massive rod for your own back, and I don't see what you gain. As others have said, alternative registries exist, and installation instructions of standalone plugins are easier on users than downloading and installing a whole editor.
Can't you just do your own "Copilot" extension and plumb in an instance to ollama? Isn't that what Continue does? https://marketplace.visualstudio.com/items?itemName=Continue...
But then they wouldn’t be able to capture as much analytics.
So early stage that we can't download it? Why would I want to join a discord server for a code editor?
What is the plan when VS Code introduces all your features?
Well the code is there: https://github.com/voideditor/void/tree/main/extensions/void But it's super basic right now... Like literally just the bare minimum to make an LLM call in a side window and a view to apply/reject diffs. Pretty much a weekend project at this point.
Maybe that's the point. Who is going to be acquired by Microsoft and then just be part of VS Code?
I wish someone would fork Omnivim2 instead in order to better compete with VS Code. The editor looks like VS Code, but it is natively compiled and supports VS Code plugins is my understanding. Unfortunately the project seems abandoned. Its coded in ReasonML.
I'm sick of dev tools being built around web browsers. We used to have fully native apps in the 90s and 00s that didn't require so much complexity compared to today.
Edit:
Forgot to mention that it also is powered by vim under the hood, but that's a strength in my eyes.
https://github.com/onivim/oni2
there is also PearAI - "The Open Source AI-Powered Code Editor. A fork of VSCode and Continue." [0]. It's getting very crowded in this space: cursor.sh, continue.dev, double.bot, supermaven, codium.ai, PearAI and now Void.
[0] - https://github.com/trypear/pearai-app
If you've tried others, would love to understand if there's anything you didn't like specifically (I'm one of the creators for https://double.bot)
I ended up actually not trying double.bot - overall I tried only cursor.sh (first with BYOK and now subscriber) and continue.dev (didn't manage to integrate own key from OpenRouter in JetBrains plugin). For me in the beginning one of the selling point is if I can bring my own key (ideally with GUI setup instead of some config file and working with OpenRouter key). Then if something is open source is also a selling point.
I would recommend update double.bot landing page to have all features shown in few gifs on the same page and 'show in action' short video' - continue.dev had actually pretty good landing page. CodeSnap is probably something similar like cursors' 'apply' but double.bot landing page feels more like documentation. Also at this stage probably worth to add additional comparison to cursor same like you have for copilot
Just added 'build a new landing page' to my to-do, appreciate the feedback!
Anything else we could do to make double.bot a more compelling option for you?
Seems to be a dupe of another Show HN post posted by the same user 4 days ago https://news.ycombinator.com/item?id=41523197
A small number of reposts is ok if an article hasn't had significant attention yet. This is in the FAQ: https://news.ycombinator.com/newsfaq.html#reposts.
It was more that I thought the prior discussion was worth linking to.
We did a smaller launch a few days ago, but we've actually made a lot of changes to our github contributing page, website, logo since then (based on the feedback). Just trying to ship fast, haha.
I only just realised you're YC-backed. Makes sense, glhf
Doesn't sound like it warrants another post.
Given that this one is getting traction, I think it indeed warranted another post.
launched twice? gotta launch as much as possible in this competitive landscape
If you're not shipping, you're slipping
And if you're slipping you can't dip in
A practice that is considered perfectly cromulent!
I'm surprised no one in targeting Jetbrains IDEs as seriously as Cursor has targeted vscode.
I'd imagine Jetbrains users are much more willing to pay, and the non-AI intelligence Webstorm has baked in save me almost as much cognitive load as Cursor does if not more compared to base vscode.
I tried Continue but it wasn't very smooth at all.
I started paying for Sourcegraph's Cody AI because of how great the integration was in JetBrains IDEs!
Happy user for a few months now :)
Try out the Claude dev vscode extension. It's much better than Continue in my experience.
Supermaven[1] has addons for vscode as well for jetbrains (and neovim). And it is so good that I quickly decided for their paid plan
[1] https://supermaven.com/
They have announced they are building an editor; so while I'm a subscriber, I'm not sure how long the plugins will last
they really did? where could I read about this.
I prefer to stay with addon
Edit: I see it, this is in their newest blog post
They announced it with their new round
https://x.com/SupermavenAI/status/1835743882971426837
Yeah, my first thought is the plugins are going away sooner or later.
I hope they use Jetbrains as a base instead of VS Code (or roll their own). I strongly prefer IdeaVim or over the one in VS Code. It's the main reason why I don't use Cursor.
There is Cody (https://sourcegraph.com/cody). But their "applying code" feature is worse than Cursor. They put import lines on the bottom of the file, for example. Cursor's "applying code" is superb. Too bad it cannot be used with another IDE.
JetBrains + CoPilot integration is pretty good.
We're targeting android studio - https://firebender.com. I imagine its because Java Swing is actually extremely painful to work with and learn. Most developers interested in AI and the experience overlap with Sun Microsystem's Java Swing is not very large.
Jetbrains isn't open source and this seems to be a fork of vscode, not an extension so there's a bit more control on what's being done here.
Edit : This is only my guess, I have no internal knowledge of Cursor.
Intellij community edition is open source, but not the enterprise version. However, CE is pretty feature rich with a ton of users and if you're targeting indie developers, most are on CE anyways. I'd say if you do Intellij, you have to be in for the long game to build up to enterprise, but its doable.
That's interesting actually, I didn't know they even had a community edition but as I am reading anyway it seem it supports only Android, Swing, and JavaFX. That's quite a limitation no?
AFAIK jetbrains has an AI Assistant now, too. But I cannot say anything about how it works.
It was sub-par when I last tried it, like many things JetBrains have been doing these days.
I love their products, and I would be sad if they can't stay competitive, but it looks like they are taking all the wrong turns.
you can try https://firebender.com. Its specifically designed for android studio, but still works well for intellij. I use it every day to develop the plugin itself.
Thanks. I'm already using GitHub CoPilot, but will give this a try.
Honestly speaking I'm not a big fan of chat based assistants, which was one of my complaints about IntelliJ's own assistant. Auto-complete feels more natural, although I can see that the suggestions might be more accurate and to the point with the chat based ones.
we're working on this right now. Expect to see auto-complete, and cmd+K functionality this week!
It's close to useless.
I would know, I made the mistake of paying yearly.
I don't know why, but I somehow expected that...
Plus points if keeping the tool free for home/OSS users and then charge for corporate or on-premise deployments.
One critical question to the users:
Is Void able to understand the source code files from the project as a whole?
What I mean is being able to read the project, and then make suggestions based on the code that was written and propose new functions? For example, I would see it as fantastic to write unit tests where I'd say: "Please write a unit test that is verifying the functions on ABC.class" or "update the unit tests for ABC.class".
I know it is difficult for an LLM to create a valid result on the first try, but at least it does same me a lot of time creating a skeleton with most of the functions and samples already there. Most times just have to fine tune or fix a few parts but can say it amplifies my output speed by at least 10x to have that kind of functionality.
What is your impression? Can it read all source code files on the project?
> All of the code is OOP-based, and they mount DOM nodes the old-school way (which is what React was supposed to solve..)
I don't know about VS Code, but I remember Atom was refactored to use manual DOM updates because the performance penalty of using React wasn't worth it.[1] By the way, isn't OOP by far the most popular paradigm for building desktop UIs? I imagine VS Code is a difficult codebase to work with that has a lot of intricate code (as is usually the case with large software projects), but that's a strange piece of criticism :-)
1. https://github.com/atom/atom/pull/5624
I have spent a great deal of time wading through the VS Code codebase and it takes OOP to the extreme. There are mile long inheritance chains, everything is a class, and it is a giant bowl of spaghetti. To some extent, I understand why they used that development approach. It doesn't use a UI framework, just DOM APIs, so classes make sense in lieu of components, but it's still bonkers.
> By the way, isn't OOP by far the most popular paradigm for building desktop UIs?
Yes, but I wish it wasn't. My day job is desktop development (with Electron), and I avoid OOP as much as I can and try to use a functional approach. After jumping all over the VS Code codebase to try to understand how some of this stuff works, and seeing how hard it is to navigate, I think heavy OOP is a bad idea.
Always interesting to hear such claims when graphic editors like Penpot, which have much tighter perf requirements than editors, are so fast while using React.
How do you implement a text editor?
I think a lot of people don't consider how complex text editors are and then there are the complexities of an IDE, or extensions, to be able to interact with it.
ropes? gap buffers? https://code.visualstudio.com/blogs/2018/03/23/text-buffer-r...
how vscode does it,
- https://code.visualstudio.com/blogs/2018/03/23/text-buffer-r...
- https://en.wikipedia.org/wiki/Piece_table
I'm not saying text editors are on the last place in terms of perf requirements.
I'm just saying, that general graphics editors is above that and do just fine with React.
This is cool! I'm a regular user of Cursor (I barely write any code now - just prompt and tab tab tab).
The thing that's really keeping me in Cursor (vs decamping for Zed) is NOT the Cmd K or the sidebar chat. It's actual Cursor's Tab autocomplete model. I've tried many other tab completion models and Cursor Tab blows them out of the water. Supermaven's new release is promising, however.
For the use-case you're solving for (issues with code-privacy), I think something like https://codeium.com/ does already allow for on-prem deployments with enterprise support. I'm trying to think of who would be served well by a fork of VSCode vs Continue.dev or something like a codeium VPS deployment.
Curious: what is your use case for cursor, i.e. what is the problem domain or complexity level (on a scale from boilerplate code to core architecture) that you typically work with? Also what plan are you on and is that enough for your needs?
Hi, we're working with an large typescript codebase and I develop fully fledged features with Cursor which involve non-trivial UI wiring up work. Decent amount of state management using React too.
It's definitely not boilerplate level code - but it also also stops well short of architecture level work. Here's a recent prompt of me trying to implement line-by-line diff generation: https://pastebin.com/WRJpNwqc
When I need to do architectural/design work, I do it in multiple passes, not generating any code, just lots detailed text with tons of feedback/back-and-forth with the model.
I'm on the pro plan but it's not enough for my needs - I run through the quota in 10-15 days. Then, I revert to using Sonnet 3.5 via API key.
Big fan of Codeium here. They're the only player who took vim/neovim users seriously from the start, and it's great to see people building extensions like monkoose/neocodeium. Their models or ux may not always be the best available, but their understanding of their target market and their product positioning is peak IMO, particularly the enterprise/on-prem options you mentioned. My money's on them long-term.
Yeah, I have not given enough time and did so yesterday; it suggests new parameters for example, so I only have to begin a "refactoring" and then tab tab... Vim guys are missing out.
My projects aren’t super complicated and just mostly Python/flask/htmx/html templates. I started with cursor during their free trial so I got the 500 sonnet prompts and it was pretty awesome. Then that ended and the remaining “free” models were fairly lacking.
So I went a’hunting, and settled on the Codeium vscode plug-in free plan which while not as good as cursor/sonnet combination but is quite capable and their “free” models are capable enough for what I am working on (believe they even use a llama 3.1 model— I think they have various models, but left it on the “base model” and it’s been pretty good)
Why is there a waiting list if it's open source?
not fully stable yet you can build it yourself from github
If anyone can point in the direction of a Cursor like experience, but within Neovim, I'm definitely in the market for that option.
- https://github.com/olimorris/codecompanion.nvim
- https://github.com/yetone/avante.nvim
Did not have the time to test them yet and there are probably more.
Hi, thanks for sharing. A question: What is the advantage of this approach compared to the tool I'm using right now - continue.dev ?
Oh, I’m yet to find a good alternative to Cursor’s RAG-powered side chat. It helps me work with huge codebases so much. Tried Continue, but it’s very unstable, and doesn’t work as well. Would prefer a command line solution, vscode plugin is the next choice, having a separate editor is not ideal, but I’m glad there’s some competition.
Aider is AI pair programming in your terminal.
Aider lets you pair program with LLMs, to edit code in your local git repository. Start a new project or work with an existing git repo. Aider works best with GPT-4o & Claude 3.5 Sonnet and can connect to almost any LLM.
https://aider.chat/
See LLM Leaderboards as well: https://aider.chat/docs/leaderboards/
Thanks! I use Aider with Claude 3.5 Sonnet on smaller projects sometimes, and it's really fun and helpful when it can put a whole repo map into the LLM's context.
> Would prefer a command line solution
I haven't tried this myself so I apologize if it ends up being bad, but I've seen Aider [0] get linked a few times from people who wanted a CLI solution for AI code completion.
[0]: https://aider.chat/
Aider is really cool for small projects, but it builds a repo map instead of using RAG. That works on small codebases, but totally fails to be useful on large ones.
How large?
In use, treesitter-derived approach seems far effective than rag.
>5k source files. They don't fit into the context. I know I can limit what is sent, and I can attach files in the Aider chat myself. But this is not perfect for making an LLM answer questions about a codebase when I don't know much context beforehand. With Cursor, I can just do "@codebase How is a %feature% implemented?", and it's very quick and often helpful with a couple of follow-ups.
Not sure if this is what you'd like, but I remember this repo from some time ago https://github.com/Storia-AI/sage
Storia-AI/sage: Chat with any codebase with 2 commands
Thanks for the suggestion! It needs some work to set up, and it looks like it only works on Github repos. Also, to work with non-local LLMs, you can only use Pinecone for vector storage. I might have misunderstood something, but I will check it out again later.
When I see "Waitlist" and "Discord" on a website, my immediate reaction is to close the tab and forget about the project.
And well, complain about it on hn obviously.
But why does it have to be a VSCode fork? Couldn’t it be a VSCode plugin?
Don’t listen to the naysayers: this can be Dropbox all over again! I love Cursor but, it’s a pity it’s not OSS.
The success of Void is tied to how well you implement the part in which AI reasons about multiple files. This is where all the sauce is in Cursor at the moment imo.
The repository contains the full source code but there's a waitlist to download? Why?
> Any LLM, Anywhere.
> Host your own models locally, or communicate directly with your favorites.
Looks fantastic, will happily try this!
Hey congrats on the release. I don't have much to add other than to ask for more context about this statement-
> allowing AI to edit multiple files at once
Cursor already modifies multiple files at the same time. I typically do this by pressing cmd+shift+I and in the left side of the window the required files will be shown as they're being modified.
> Microsoft also made its extension marketplace closed-source so we (and Cursor) have to hack our way through it.
For the marketplace, can you not use Open VSX [1]?
[1] https://github.com/eclipse/openvsx
I want to build my own agents so I can my private domain specific awareness in the coding environment: PRDs, product docs, API docs, business goals, etc. I'm assuming I could point this to any API compatible endpoint and build my own agents?
Tried the get in touch link, but the Google form says I need permission to view it.
Should be fixed now!
Why is this a Show HN not a launch HN? Is it simply a concept and not yet fleshed out?
Hard sell to choose you especially as a full on editor replacement (as well as a copilot replacement) without a demo or even gifs.
Waitlist.
You mention refactoring. Do you hope to be able to merge new VScode improvements as they arise, or is this a hard fork that will diverge from VSCode from now on?
Cursor needs some competition. Their generation quality fell off a cliff recently as they are doing behind the scenes product optimizations with LLMs
I'm not sure why SourceGraph is struggling so much that I have to do it for them (maybe the name?), but Cody is competition and has the benefit of being an extension instead of a fork.
It’s just been so long since going from ‘head to code’ has been the hard part for me. I’d imagine you have to be very unfamiliar for this to be useful.
My primary usage of cursor is cmd+L and cmd+K, I'll give it a shot. Also, makes sense to make an open source alternative. How do you plan on monetizing?
We plan on selling to enterprise companies. We know a lot of companies want to keep their repos private and so we offer them on-prem hosting options. Closed source tools can't really do this because they have to keep their prompt building algorithms behind a private API that they own.
This seems to work pretty well for Codeium, which offers its product free to consumers (at a loss), and is still able to make net profit from enterprise companies. At least that's the plan for now!
>We know a lot of companies want to keep their repos private and so we offer them on-prem hosting options.
Is there anything more than this?
makes sense! Excited for y'all
Is this really "open source"? Can I download all source code and run it locally without internet connection?
If I create an MIT licensed RSS reader, distribute it on GitHub for everyone.
That's open source, just because you need an internet connection doesn't preclude the 'open-source'-ness right.
Sir, this is a text editor.
I do understand your frustration but again to be pedantic, the type of software isn't conditional to how it's licensing will work.
I think there is a good discussion to be had over the licensing of LLM's and the fruits of training data, as well as that of the training data itself, but when it comes to the software itself, it's hard to argue what is and isn't open source.
> Can I download all source code and run it locally without internet connection?
This is not what open source means.
Well, of course, open source would mean that you can also download all the training data and re-do the training yourself.
> open source would mean that you can also download all the training data and re-do the training yourself
I don't think so.. Utilizing another data source, does not mean they're required, or even permitted in a lot of cases, to distribute that with their software/service.
Genuine question: why don’t you started from VSCodium if the the VSCode is too much hassle to deal with?
Isn't VSCodium built from the same source code as VSCode? Excluding a few proprietary things Microsoft puts into the official binaries.
Great job! Can you share where the codebase prompts are listed? Maybe it is a valuable thing to learn
Why the completely generic name, with meaning unrelated in any way to editors?
That's intriguing. I'd like to be a user.
Honestly, I'd rather have a buggish extension than having to change editor.
> we're building Void as a fork of vscode
Hard pass.
Sorry but you can pry IDEA from my cold dead hands or at least VSCode doesn't hold a candle to it IMHO. And no, I'm not looking for a discussion on VSCode vs IDEA.
Building a new code editor for a single feature is a dead end in my opinion.
does it support, ssh workspaces like vs code?
does it support, ssh workspaces like vs code
most exciting is that it will run locally and we don't have to use middleman and give up our juicy data. lol
Ughh. I'm tired of these open source vc backed companies. Can't wait for this meta to die out.
This investment by YC in both Void and Continue.dev perfectly exemplifies the VC spray-and-pray approach by backing two nearly identical open-source VSCode plugins for AI-assisted coding.
They're simply throwing money at similar projects in the same space, hoping one will stick, without any special insight or reason why one should be more successful over another.
This is VC operations by design since forever.
Heck, even centuries ago, they'd fund multiple ship crews to navigate the Earth, hoping one would find smth interesting.
It's the model that works in high-risk, high-reward endeavors.
I’m interested; are you suggested this is a bad thing, or just highlighting something of interest?
Not a bad thing, but just a reflection that among some of these products there is nothing special when compared to rest of the bunch, there are very similar products (even backed by the same investor), and us users have no clear idea what is different between them, if anything, and why we should use one over another.
I guess similar theme can be seen in LLM/AI space in general, or previously in the streaming services, just throw as much money at the problem in the hope that your solution will stick, without any significant and clear differentiation.
double.bot also backed by YC
> This investment by YC in both Void and Continue.dev
Oh there's a lot more than that.
Just the two that I have recently been checking out and, as many other commenters in this thread asked, what is the difference between the two? I guess investors know something we don't, or they are just throwing money at the wall...
release date?
Why are so many coding assistant forks? Is it fashion or is there a technical reason? Imagine Python support being a fork. Or live server.
Because it's hard to reimplement VSCode from scratch?
VSCode has plugins
They are making changes that cannot be done or are very difficult with vscode extensions
What might they be? API calls, read all files, write to files, side bar, UI interactions in editor. These are things I would imagine an extension can do. Continue.dev is an example.