Launch HN: Haystack (YC S24) – Visualize and edit code on an infinite canvas (github.com)

328 points by akshaysg ↗ HN
Hi HN, we’re Akshay and Jake from Haystack (https://haystackeditor.com/). Haystack re-imagines the IDE by putting it on a 2D digital whiteboard that automatically draws connections between code as you navigate and edit files. We designed it to match our mental model of how code works – as a graph of connected components. Here’s a demo of what it looks like: https://www.youtube.com/watch?v=5XADctpWNNs. You can download Haystack from https://haystackeditor.com/.

We used to work as software engineers at big companies, and when the explosion in AI code-generation tools started we were really excited about them. They gave us a huge speed boost in our day-to-day work. But something was missing: writing code had become easier, but navigating codebases remained painful. We found that we were spending more time on the latter than the former!

Our vision of Haystack is an IDE that helps you navigate your codebase at speed. You should be able to type a vague description of the code you’re looking for (“Show me the code flow that triggers after we click the submit button”) and immediately see the right result. But we don’t think the traditional split-panel interface is the best fit for this type of rapid navigation. Inspired by the fluidity of design tools like Figma (Akshay used to work there!) and Miro, we realized that the canvas layout is a perfect way to represent and operate on code flows, and a great base for all the AI and collaborative tooling we want to build! This is how Haystack was born.

You might have seen our original Show HN (https://news.ycombinator.com/item?id=41068719) about two months ago. Since then, a ton of our users have told us that they’d love to contribute. We’ve made the decision to make the editor source-available so that folks can make contributions and so that they can examine the code to make sure they can trust it. Our repo is at https://github.com/haystackeditor/haystack-editor.

If you’re wondering why we’re going source-available rather than proper open-source: as a startup, we're going to eventually monetize and it’s too early for us to make the decision on whether going fully open-source is the right move. We want to own the distribution of Haystack until we can better understand the ramifications that open-sourcing would have, and whether it’s sustainable business-wise. We’d love to find a way to make proper open-source work, but it’s a one-way door, so we want to take our time to make sure we’re making an informed decision.

We would love to hear about what you think about having a canvas in your IDE and the role visualizations have to play in software development!

175 comments

[ 3.8 ms ] story [ 208 ms ] thread
This looks quite useful! However could Haystack be made an VSCode extension instead of a fork? So I could use this in Cursor, which is also a VSCode fork.
We're going to make a "Haystack-lite" extension in the future! I understand the pain here. We chose to go with a VS-code fork so we could maximize the canvas features. I am curious if you would find the "canvas" view distracting when it's on a sidebar as opposed to the main editor?
Thinking about it, a sidebar "canvas" could be more useful comparing to the main editor to me:

* when reading code, I found I jump back and forth the call stack quite often, a visualization of this could help with this navigation, especially with some properly designed shortcut keys.

* I mostly code on MacBook so screen real estate is precious. The canvas as main editor looks like waste a lot of screen space. But the canvas in a sidebar do not have this issue.

1. You can hop back and forth using the "backward/forward" buttons in the top bar, similar to VS Code. Not sure if I misunderstood here. 2. That's fair. You can "pin" editors on the canvas, which allows you to fullscreen editors.
I think part of what makes this work is that ultrawidescreen monitors aren't as impossibly expensive as they once were, so screen real estate can be spent and your spacial reasoning can take over. It might just not work well on a laptop screen.
So many YC folks are forking VSCode for various reasons that mostly revolve around "can't monetize extensions, want to own the platform".

I don't think any of them will be successful, IMO. You want to be an extension because getting software approved is _hard_ at bigcos, it's much easier to trojan horse on an existing tool.

I could be wrong but I don't believe it's difficult to monetize an extension.

The reason I think there has been an explosion of VS Code forks (e.g. Supermaven, a very successful extension) is that being an extension in VS Code is limiting insofar as what you're able to change in terms of the UI and UX.

You are right about the difficulty of getting into big companies. However, we developed a standalone editor because it's easier to build on top of, and we eventually want to build a very portable browser-based editor that utilizes the canvas view for pull requests, arbitrary code, etc.

Bigcos not that hard to get into. It's just an entirely different process. You would need to hire someone with that kind of experience.
I think Cursor is already looking successful. Most VS Code people I know (including myself) have switched over. I love the idea of Haystack (I was a fan of Light Table back in the day) but if I have to pick between a better UI and an assistant who perform most simple code transforms by English language request, saving hour(s) per day, I’m picking the assistant.
Yeah the productivity optimization here makes sense. We intend to add generative AI features unique to the canvas UI as well to ensure our users don't lose productivity.

I am curious what features you like about Cursor the most? For me it's the CMD/CTRL+K -- I've had mixed experiences with the chat window and Composer.

Cmd-K and tab complete with or without writing some guiding comments
(comment deleted)
I could be wrong but one of the main reasons to fork VS Code is because extensions have a limited UI.

This was at least the motivation for Positron, AFAIK.

This feels helpful as I have to hold less stuff in my mind, but the UI is way too mouse dependent right now. You guys should make some shortcuts and somehow make it “tiling” like a window manager (dwm), so you don’t have to manually resize windows.
We actually do try to make the canvas as keyboard navigable as possible. There are a list of keyboard bindings in the "Keybindings and Help" button on the bottom right.

In terms of manually resizing, we actually do try to tile in the viewport. May I know what's causing you to resize it so I can fix it?

In this vein, do you know what the experience has been like for users that use the vim vscode plugin? I see that it's trivial to carry over extensions and whatnot, but is the experience still "good" trying to use both?
I use the Vim VS Code plugin! It mostly works, but I'm not going to lie: there are a couple of bugs e.g. when you go to definition the Vim selection gets confused because in most cases you get placed into a symbol editor.

I am happy to fix any bugs here. I am the only person I know who uses the Vim plugin in Haystack, so I haven't prioritized fixing the bugs I encounter.

Sorry, I was just going off the demo. Maybe demo the keybinds briefly.

If you get a demo of someone editing stuff super quickly using only keybinds I think it would look pretty cool.

Right now I don’t know if the value prop is strong enough for me to switch editors and remodel/relearn all my keybinds. This is like a 10-15% improvement not a 1.5x or 2x improvement.

An extension would probably help onboard users.

Interesting, for me this is exactly the opposite of efficient code navigation.

LSP + keyboard-driven interface is miles ahead in terms of efficiency and speed.

I respect the effort of breaking into less researched lands though. Congrats on the launch, I'm sure some will find this tool useful.

This is fair! Not sure if you're a VS Code user, but our goal is to maintain a parity between the keyboard experience in VS Code and the keyboard experience in Haystack. All the keyboard shortcuts in VS Code still live in Haystack.

If you're a Vim user then this is an entirely different story haha.

To give you (Haystack authors) another datapoint, I'm a +1 to this comment. I can't think of a time when a canvas UI solved a problem I had. Usually it made navigation more difficult.

But also agree that it's great that you're innovating in this space! Good luck!

I actually like keyboard-driven interface + spatial layout, I just don't have a large enough monitor (I've been hoping for 40"+ 8k monitors for years). But having multiple files open that smartly refocus and rearrange as you ESC-. (Emacs shortcut to jump to implementation of a thing under cursor) would be wonderful: if smart "arrows" are drawn between file boxes, that'd be sweet as well.

The hard bit is making that to be "smart" and not annoying with things moving that you still need.

Is this a YC backed company?
Yes (all launch HNs are YC-backed)
Congrats on the launch. It is an “obvious” idea, in a good way. Agree that minimizing the effort in constructing the workspace will be key. I’d also love to try a sort-of 3D stack for caller/callee relationship, zooming in and out. The code really is a graph (not acyclic though), and I think visualizing it that way can be helpful.
I'd love to learn more about what you're envisioning for the caller/callee relationship here! Do you mean recursively show callers/callees as you zoom out?
hey love the concept (though Im not sure if its for me).

btw you should add a (small) gif or video in the github readme cuz that makes it really clear of what exactly its about.

I get that this is a completely new idea so adoption is going to be slow but I'm curious to see how your user interviews and user research is going. I'm a young dev but I'm not even sure if I would use this because the way I've learned to code is already so ingrained into the traditional IDE. Nevertheless, congrats on the launch!
Jake from Haystack here! Thanks for the interest. We've been doing a lot of user interviews with folks on our discord (and folks we know in real life) to understand issues in the onboarding process, find problems that make people churn off, etc with the long-term goal of increasing retention. A lot of our time lately has been spent on fixing these bugs and improving the new-user experience. As a result of this we found that retention on later launches got much better!
Have you done user research outside your fan base to show this experience is something developers want?

Is it desirable enough to get them to switch IDEs?

Yeah just to add onto what Jake said -- in our user interviews we found that folks, even those who are pretty used to old IDEs, do get the hang of Haystack pretty fast. However, developers are very idiosyncratic, so we frequently get pain points that are real problems but are also highly unique to each developer, which is interesting!
Totally: I couldn't stop thinking about actually generalizing this to a workspace/desktop so I could apply similar principles with multi-windowed Emacs — a window manager that also connects windows ;)

Looks really intriguing, but Emacs is really hard to leave behind.

But I do believe this pattern applies to more things than just IDEs, and might even replace multi-workspace setup. Keep at it and hopefully you are successful and this pattern keeps evolving!

This is really, really cool!

On the licensing side, though:

> We’ve made the decision to make the editor source-available so that folks can make contributions and so that they can examine the code to make sure they can trust it.

This is a very reasonable intent, and it absolutely makes sense to preserve your options to monetize, but Polyform Strict https://polyformproject.org/licenses/strict/1.0.0/ , which you link to from your readme, seems to be far stronger than this intent.

The way I read that license, I cannot "make changes" even if I do not distribute those changes, not even changing a line of code on a version I'm running on my own computer, whether I'm using it for a commercial project or not. So contributors, it seems, would be limited to raising issues but not PRs? And it's unclear what applies as commercial use - if I'm using Haystack on a commercial codebase, and I encounter a bug that triggers on that codebase, and want to use the Haystack source code to understand what's going on to further my commercial purpose, would that be an unlicensed use of the source code?

Now, make no mistake - this is in no ways worse than the prior status quo of Haystack being closed-source. And I think you as a company should focus on building, with the certainty and comfort that this license, if anything, does err on the side of being restrictive. But some users, given this license, will be uncomfortable referring to the repository at all. I'll be one of those, but will still be excited to experiment with Haystack as a binary distribution nonetheless!

(Not a lawyer, this is not legal advice!)

Hmm I'll take a look at other licenses to see if there's one that expresses our intent in clearer language.

The intent is that you should be able to make PRs, examine the source code, but not distribute Haystack for commercial reasons.

I believe the license should allow you make PRs and examine code (even in commercial settings), but I'll take a more in-depth look.

Thank you for raising this to my attention!

Very cool. I imagined my organizations entire codebase being mapped like this and across different frameworks and languages.

I don’t know if I missed this in the video or if it’s not yet possible, but that’s a lot of manual work, so instead of connecting the nodes, give a simple bot to run in the repo folder to automate the visualization.

It’s super cool and I’m adding to my watch list.

If I were you, i’d target enterprise organizations or local municipalities IT groups who are going through or planning their digital transformations. They have a need for way-finding and sense making across legacy and new work.

If you play your cards right, Salesforce will come knocking in under a year. I see a lot of compatibility with their vision and product offerings.

It automatically adds connections as you traverse through files/symbols. Not super sure if this is what you mean, but we intend to serve the users "flows" based on queries in the natural language e.g. "what happens when you click the subscribe button from the mouse down all the way to the backend".
That’s awesome! Generally when I find tools that can improve our work culture and output velocity and quality, I try and get it on front of teams ahead of a an internal hackathon where the cost of on-boarding and prototyping is already factored in. This feels like a good candidate.
that would be interesting if you could "link" across API calls to different services.
Most likely we would either have to get the user to do this (we plan to add this as a feature) or use AI to understand when disconnected code (from the LSP standpoint) is indeed connected.

This is a top user request though!

I like the concept so far in the video. I just want more, even bigger, more panes, all connected! If it's usable in such a state I don't know, but would be fun to try. Like a UML diagrams of the whole code base with thousands of arrows.

Ane I can then tell my coworker that the file needing change is in the upper right corners somewhere.

And git diffs being visual, can just zoom in on the changes here and there.

What happens when I have 10,000 or so editors open at once?
I've actually never tried this (it'll probably take awhile to get to this point). Feel free to try it out and report back the results -- I'm curious as well!
This has been a great idea for decades. I want Haystack to be successful just like many other attempts. The early execution seems promising. And I suspect there will be many challenges (e.g. when it's hard to figure out caller/callee,, inconsistent UX preferences across developers, etc). Kudos for taking this on!

Btw I've always thought that this is even more powerful when the screen estate is more infinite than a 2D screen (like in a VR headset).

I love the idea of a Haystack VR world! It's a shame that VR software is in a tenuous state due to the biological factors, but I believe it's the future "one day".
"In a tenuous state due to the biological factors" is easily the funniest SV euphemism for "can't be used by humans."
It’s ok after they deliver the MVP there will be a wetware update.
Who knows, one day it may be possible (hopefully without any dystopian updates to human biology)!
At this point, that seems dubious - your inner ear is going to go all inner ear on you, no matter what. Unless you get to turn that off, VR is not it.
I hope our code-editor-in-VR wouldn't involve flying around like hilarious depictions of "The Gibson" in bad sci-fi movies.
Doesn't matter. As long as you use VR to display a virtual 3D environment and you move within it, your inner ear will fight with your vision system if you're moving or not. If the visual system and the accelerometer don't agree, the positioning system throws an exception.

And, for whatever reason, the human exception handler for that problem is firmly linked to the barf() subroutine ;)

Like I said, as long as you're not flying around- moving within it, then your inner ear doesn't care. Turning your head doesn't count. I don't see a need in a code-in-VR system to move like that. And most VR games solve this my having your teleport instead of translate.

I think the barf routine is because when your brain senses your vestibular system not working it thinks "oops I must be poisoned" and tries to make you throw up.

"Fixing our eyeballs is just an engineering problem."
Is most development of Haystack done using Haystack?
Yes! Both of us prefer Haystack to other IDEs; for me it feels a lot less "restricting" than VSCode. Developing in Haystack also helps us find issues early.
Does this project take any inspiration from Light Table?

Potential feature request - take the visualization beyond 2D. Really complex systems are usually an intricate graph (polite way of saying ball of yarn) and just visualizing in the 2D plane doesn't seem to cut it in my experience.

I would love to have something like concentric spheres of visualization - boundary services on the outer layer, core/domain services in the inner layer.

A lot of people mentioned Light Table, but I didn't know about it until post-launch!

It's interesting that you mention a more 3D view. We were ideating on this, but it's really hard to get right in a way that the user can understand. This is definitely part of our vision for the future though!

EDIT: Deleted the bit about my speculation regarding Light Table since it's false.

Light Table was getting lift and traction but didn't make it. There was a ton of conversation on it back in the day on HN, including great presentations by the creator (who had been a PM on Visual Studio as MSFT!). Definitely check it out.

There's definitely demand for this tool but it looks like it's hard to get it off the ground. Wishing you the best of luck!

someone should revive it as a GUI debugger. doing so would narrow the scope which should cut down on complexity. let people use their own editors and IDEs for making code changes. the focus should be on inspecting state, seeing live updates, allowing you to smoothly change inputs and immediately see the output, plus the usual debugger features like stepping through the program.
For 3D I’d run a clustering algo across the graph and use the Z axis to background “far” clusters.

I’m imagining a typical graph might end up looking a bit like this: https://upload.wikimedia.org/wikipedia/commons/thumb/8/81/07... in which case I’d want to have whatever cluster/file/method/object I’m focused on front and center with some visual cues like low-contrast and z-axis perspective that preserves edges to show the “far” clusters that interact with the focal point.

edit to add: as per usual, naming the clusters automatically might be hard.

Interesting idea and curious - because when I look at that visualization, the first thing it reminds me of is gource.
I had not heard of Gource, and it looks like it’s building its graph based on directory layout rather than code semantics but yep that’s about where I would start!
Not just Light Table from 2010s, but also I would expect that the authors know about Oberon from late 1980s. It also features code on a boundless canvas.
Very awesome!

I want this for C# and Python! Are their any technical challenges there? I would love to use this with Unity.

For C# unfortunately we're missing a lot of features because the VS Code ecosystem itself lacks them (I think this is due to them being unique to Visual Studio). In most cases, except grabbing incoming/outgoing calls, Haystack should just work.

For Python, it should work just fine!

Can I still do the cool visualization ?

I'll experiment with this over the weekend and I might share it with my game dev friends!

Yes the visualization should work!
There are multiple ways to go about this in .NET - the instrumentation for the networking stack is not missing:

https://learn.microsoft.com/en-us/dotnet/fundamentals/networ...

https://learn.microsoft.com/en-us/dotnet/fundamentals/networ...

It is, however, not dependent on VS Code and exists standalone, so it requires integration work if there's custom tooling at play. It is what most vendor SDKs that offer observability do.

There are CLI tools which consume the EventSource events and the EventPipe itself which can be used to access them in and out of process: https://learn.microsoft.com/en-us/dotnet/core/diagnostics/ev...

EventPipe/EventSource and DiagnosticPort are very powerful APIs however that cover much deeper instrumentation scenarios, so they are going to require writing a connector for the product that wants to integrate with them. There is already an implementation in shared code that is used by e.g. dotnet-trace which can be found here https://github.com/dotnet/diagnostics/tree/main/src/Microsof... but it can be a bit convoluted to read through.

This reminds me of the "Kansas" environment in the Self programming language. Here is a video about it from Sun in 1995:

https://www.youtube.com/watch?v=Ox5P7QyL774

More interesting search features were added after this video was made.

> Haystack takes care of the tedious confusing parts of coding

Tedious, yes, but confusing? Yikes!

My job as a coder is to resolve ambiguity favorably. Every day, I strive to leave the world less confusing than I found it. I would no more rely on an IDE to "take care" of confusion than I would outsource my firm's core competencies.

But the main conceit is cool, and I'm sure the dependency graph is helpful. I'm just picking nits with marketing copy. Can't wait to check it out! (Although my first instinct is to check for a canvas plugin for VScode).

Edit: I should say that I think this UI paradigm will fundamentally improve coder understanding by exploiting our visual cortex and "sixth sense" of proprioception, at least as far as I understand it via Supersizing the Mind (Andy Clark, 2008).

Hmm I think someone else has voiced a similar opinion on the wording here, and I actually do agree with you.

What was meant by this statement is really just taking care of plumbing and the visualization of how code relates to one another.

To give you a very basic/silly example of plumbing, if I add a parameter in a function used in 100s of places, doing the manual work to figure out what edits I need to make to those callers is pretty annoying and in most cases mechanical.

If you're lucky, you can do a grep + replace (or just use a default param), but in most cases I find it requires more manual intervention than that.

Adding a parameter to the function signature, and then passing an additional value at each call site is done... automatically in Haystack? Impossible!

Edit: And "to figure out what edits I need to make" is done by any ol' compiler, no?

Not yet -- it's our vision to automate stuff like that!

> Edit: And "to figure out what edits I need to make" is done by any ol' compiler, no?

Kinda? There are cases where you have to drill data down into the callstack (i.e. pass data from above). There are folks who have gone down this path before: https://www.jetbrains.com/help/resharper/Refactorings__Chang... but we intend to automate it even more and show it to the user in an organized way for them to handle large-scale "refactors".

I strive for the same thing, but every time surprising new requirements arrive, there's a good chance that the elegant architecture my team created is in conflict with the design we now want. At that point, there is a three step process:

1. Design the new architecture.

2. Gain a full understanding of the existing architecture and all of its trade-offs.

3. Design a way to convert the old architecture into the new architecture.

This happens surprisingly often and it's quite unavoidable. In the past, I tried to avoid redesigning by building things that could be expanded in every conceivable way, but I discovered that over-engineering actually makes it harder to redesign. YAGNI wins!

I've been thinking for a while that a canvas-type editor might help a lot in the process of redesigning. It should help by taking better advantage of spatial memory.

So this is a major side question, but is there a go-to open source infinite canvas? I'm building a card game and I need a multiplayer infinite canvas for it. I'd appreciate any recommendations.
Does tldraw work for you? We use Pixi JS, which comes with graphics as well.
Ooh I see PixiJs can make content available to the screenreader for a11y. That’s always a big question mark for me when people start cramming a content tree into a canvas.
Yes a11y is very important! I do think this creates an HTML element per sprite/graphic, but the slowdown should not be terrible.
Last I checked Miro also uses Pixi so that's another quality infinite canvas app.
I think the biggest pain point for me is that nothing I've found supports cards that can be "flipped" so that they are face up or face down. I'm not an advanced coder but maybe I'll try to whip something up.
> we're going to eventually monetize

How do you plan on monetizing?

Collaborative and generative AI features. E.g. sharing Haystack workspaces, searching your codebase in the natural language, making edits to multiple files, etc.
> Monetization

All right

> Collaborative

OH! YES!

> and generative AI features

Oh no.

Since it's opt-in, the generative AI features won't ever bug you unless you want to use them!
Why can't this be an extension? I love the idea and can perfectly imagine this inside a tab of the editor area. This would allow for multiple haystack tabs as well, so switching contexts would be seven easier. Having to install another IDE (even if it is based on vscode) is a bit of a bummer for me. None the less, keep it up!
VS Code extensions are pretty limited - you can override the renderer for an editor tab, but if you want multiple things within a custom interface to be treated as first-class editor tabs by the larger environment (and by the larger extension ecosystem), that's not really possible. Forks are a reasonable step as long as they don't fall behind!
Do you really need a first-class-editor for working directly from the canvas? Wouldn't it be enough to use the canvas just for navigation and show a rendered picture of the code, but open a separate editor (maybe side by side?) for the real work?
My understanding is that Haystack's mission is to explore this design space more fully than others have before, so I think they're making the right decision to be audacious here!
I understand the frustration here! As the other commenter noted, it's quite bit tougher to make Haystack as good in the form of an extension, but we do plan to eventually explore this path!
Since extensions are limited, a nice workaround imo would be to be able to turn canvas view on/off. I'm going to want to go between the canvas and the standard VSCode view, and so if I can do that within Haystack I don't need to have both installed.
If you structure your directories sensibly, you can get the same effect/value prop in every IDE and dev env.

Or is there a deeper experience/knowledge that gets unlocked coding in this paradigm?

Not sure if I completely grasp what you're saying, but you're saying that if you plan the codebase well then dependencies between files/bits of code are more easily followed?

For our users, Haystack has been helpful to understand how code works together, from simple React components to legacy C codebases that are a decade or older. I think the value prop shines when you're trying to understand how multiple pieces of code together, and this will become more useful as folks increasingly lean on AI tools to modify/create/delete multiple files.

Did you really have to pick the same name as the Haystack open source AI framework? https://haystack.deepset.ai/ https://github.com/deepset-ai/haystack

It's a very active project and it's confusing to have two projects with the same name. Besides, I don't understand why you'd give a "2D digital whiteboard that automatically draws connections between code as you navigate and edit files" the name haystack.

This is an unfortunate name collision, but in our defense the concept of a "needle in a haystack" is hardly unique and I believe a code editor is different enough from the AI framework that it's OK.

The idea is that a codebase is much like a haystack and finding the relevant portions of it for a task or specific flow would be equivalent to finding needles.

Is this just a pretty picture of the AST that I"m already navigating with go-to definition, fuzzy search, and harpoon in 2024?
Harpoon is kind of like a bookmarking tool right? I think the value prop is that you have to do less manual window management, can see connections between code better, and can context switch between tasks.

Reposting this from another comment: For our users, Haystack has been helpful to understand how code works together, from simple React components to legacy C codebases that are a decade or older. I think the value prop shines when you're trying to understand how multiple pieces of code together, and this will become more useful as folks increasingly lean on AI tools to modify/create/delete multiple files.

This is so. fucking. sexy.

Wow.

Thank you for sharing your work.

Please don't go down the GenAI bullshit bingo VC route.

Oh, this is great. And pretty much the process of what I do when I look at larger codebases, except not as clean and organized. I would normally have a half dozen terminals open into various source files at once trying to build the map in my head. Though it eventually clicks and happens, it's laborious, but it seems like something like this would help immensely. Now, the only thing left is to combine it with an ultrawide monitor.
A gulpfile... has been while since I've seen one of those.
Yeah VS Code is built on top of some older frameworks and ideas unfortunately.