A really great place to reach for this is Elixir/Phoenix now that the set theoretic type system is in place with LiveView. I have been doing some pretty large/complex codebases gradually working towards fully integrating working this way for the past 5 years and it has been great.
The general challenge becomes integrating the fractional front-end code written in your backend-language that compiles to JS with the rest of the JS ecosystem.
JaneStreet have a love of writing their own stuff from scratch so it doesn't apply to them but it might to you.
Hence most people end up with frontend-as-backend rather than backend-as-frontend.
Clojurescript not ClojureJs, and Typescript doesn't run in the browser, it has the exact same issues/limitations: it must be compiled to JavaScript first.
absolutely nobody used those outside of a language demo. We also got --erasableSyntaxOnly last year to remove them from the language. The TS team considers them “mistakes“ for a really long time now
Thanks for the correction! It was a long time ago that I used TypeScript, and making things worse (for my memory) it was an Angular project.
So today, most people and teams avoid the parts of TS that require compilation into different JS, and --erasableSyntaxOnly lets you enforce this, right?
> Is there some limitation that WASM has that JS doesn't?
You need a JS trampoline to call your WASM and make browser primitives available to it, and IIRC calls into browser code incur some extra overhead, but those are pretty manageable.
Additionally: for high level languages, source code is _much_ smaller than compiled binaries. If your initial needs are simple, your users are likely downloading more than 10x as much code.
WASM does not (yet) have access to the DOM or web APIs, meaning every direct interaction with the browser must first go through JS interop anyways, with a resulting performance penalty. There are proposals, and maybe we'll see it happen by 2035, but as of now WASM is best suited for heavier workloads where the edge in application performance outweighs the cost of JS interop with the browser rather than being a universal solution.
A framing from that podcast that helped me better understand the role of Bonsai is that it's really a framework for building incremental distributed state machines.
This framing is also in the post. From the readme:
> Bonsai itself -- this library -- is actually more generic than the above makes it sound. It allows you to build general-purpose incremental, composable state machines. Bonsai_web builds on top of that core library, specializing it for interactive browser-based UIs, but we also have Bonsai_term for building interactive terminal-based UIs.
Curious how this compares to Melange which is used by Ocaml shops as well to double up on Ocaml for both front and backend (ahrefs being the major user and sponsor). Does this mean giving up a lot of the JS ecosystem (React, graphql, etc)?
My understanding is that js_of_ocaml has been around longer and likely has a stronger ecosystem, but Melange was designed more specifically to play nice with the JS ecosystem.
FWIW if you listen to Signals and Threads (the Jane Street podcast) they talk about this.
They have dedicated UX designers who do relatively little code but focus entirely on making the UX as optimal for the traders as possible and one of them was on an episode.
They talked about how most of the normal UX rules for SW go out the window because quants and traders have such different needs than "normal software".
This matters. Beyond performance (a universal good), the definition of "good" UX is strictly dependent on the user. Fintech platforms for traders who open a multi-window thick client application(s) and spend hours in it every day have little in common with consumer mobile apps designed to hold your attention for maybe a minute. It's mind-bogglingly lazy to choose something like MUI (Material UI) as if it were some kind of reasonable foundation irrespective of context.
Is there any professional software that wouldn't benefit from high information density? Most software design is optimized for the non-users who have to approve it.
You can have high information density and still not have words running into each-other. Taking all of the space out from around text and information graphics makes them less legible, even for experienced "power users".
I think the goal here is more an aesthetic of being for "serious traders only" rather than a UI optimized for speed and clarity. The UI designers probably have little choice but to match the preferences of the user group though, even when those preferences are empirically detrimental.
"non-users" are also big targets for software. If you run a software business, chances are you want people who've never used your software to start using your software. In that case, discoverability trumps high information density.
It only adds discoverability for the things that fit the screen. In other words, you have to make it simple. I.e. the lazy way to make something "user-friendly", do as little as possible and force the user to adapt to your ways instead.
this - most of the internal software at my job that's actually important is very dense, with keyboard navigation and shortcuts. The ones that somebody makes to sell themselves to upper management while angling for a promotion look very pretty, and never get used.
And yet, we make such beautiful and vacant software ourselves for our clients, because again, its b2b, so you aren't selling to the people who actually have to use the software. Im glad at least we arent forced that often to use specific internal tools, the better ones just get adopted organically
A lot is said in this thread about user preferences for info density, but it is really a lot more than preference. there's also a big component of it that is user productivity focused. Info-dense UIs simply require a lot less scrolling around and pagination to look at relevant data. For example the ability to look at multiple charts side by side instead of having to scroll one chart after another into position enables users to spot differences that they wouldn't otherwise see.
This makes me think about the bizarre UI regressions in macOS in the last few years. I’ve got this giant, beautiful display and somehow it becomes less rather than more useful over time due to the software it’s presenting.
Jane Street made it's way into today's Wall St Journal for its quantitative prowess and hiring of Math PhD's. What was stunning to hear was that their 2nd quarter profits were a record $10.3 billion, nearly double that of Goldman Sachs and Morgan Stanley (much larger firms).
They're a trading firm that hires really clever people to use math to trade. Their strategies are, presumably, very complicated and ever changing, and secrecy about them is the point - you get bigger margins for longer if people don't learn about your 'one weird trick' (that and also that there is presumably another side to your trades losing out on upside, and they might not be best pleased to learn about it.
Can someone who understands web UI programming tell me if this would be good for my local agent to use to produce HTML based reports and outputs for me? Or for TUI outputs?
The main thing that Bonsai gets you is a nice programming model for making your UIs interactive. If you're mostly producing static reports (with maybe some basic interactive visualization widgets or something), it will not get you much over generating some variant on HTML + JS.
It makes more sense for applications where you have complex interactive logic in the UI, with a lot of different components that need to control and display some shared state.
There was this submission two months ago: https://news.ycombinator.com/item?id=48365904 . Basically they have a TUI frontend for the same thing. And the blog post says: "Bonsai_term would feel familiar to anyone who’d ever done web development here, and it had the huge advantage of being especially amenable to AI assistance. It was actually somewhat of a mystery to us how good the models were at writing Bonsai_term code, given how relatively obscure it is [...]"
i maintain a codebase that uses Bonzai (https://github.com/hazelgrove/hazel/); agent based development works just fine, though probably a bit more token churn
There's a full terminal implementation of Bonsai as well. I actually use it in my personal workflow these days to manage my contacts database! https://anil.recoil.org/notes/aoah-2025-9
Custom infrastructure; I use my site as a playground for various things over the years. It ran as the first MirageOS unikernel for many years, and recently it's been running on a zero-allocation OxCaml webserver https://anil.recoil.org/notes/oxcaml-httpz to help me get to grips with the new extensions.
Jsoo supports self-recursive and mutually recursive tail calls, but for general tail calls it would need to trampoline every single call, since the only major browser with proper tail calls is Safari.
Pretty much. UI libraries are more or less meaningless now, and I say this as someone who spent the last decade building them. Given a proper data API to work against, you're a single prompt away from a fully tested bespoke lib that's more maintainable and has no dependencies. The weeks of work you saved by reaching for one before is gone, and the tradeoff you had to make of dealing with upstream changes and lack of customization just isn't there anymore.
This is great. It focuses on utility and information density over design. It looks like someone took a terminal UI and transplanted it to the web, Bloomberg terminal style.
I'm pretty sure you can build tools with this that are fast and pleasant to use.
At Jane Street we're super excited by Functional programming and by CAML in particular, so when we need low latency software, we use OCAML, when we need hardware, we write out own langauge - HardCAML, and when we need a Web UI, we build a Web UI framework in CAML. Because we fucking love CAML.
But AI do benefit a lot from a strict compiler, and having a simple language improve a lot on type-safety, so I don't think it is to throw, even today.
JaneStreet has enough free cash to not worry about those issues. If one of their key persons dies in service, they can go out and contract the worlds leading expert in that domain, and their annualised rate isn’t even a rounding error on their bottom line.
It looks like a nice little library; but oh boy must this be so limiting for the product teams that are forced to use. Everything looks like it's straight out of the 1990s.
It might be a very performant UI framework in OCaml, but all the UI elements look extremely unpolished to me, like what a really smart high-school kid would build the GUI for his side project on Windows in 90s.
Thanks, I can use JS as a functional programming language.
Windows and especially Macintosh UIs from the late 90s/early 00s were by far the most functional UIs we've ever had since the dawn of the desktop era - user research was a serious discipline, accessibility was baked in to the frameworks, and the HID guidelines that devs adhered to gave us uniform conventions that we users could rely on.
If the biggest criticism of Bonsai is "it ain't pretty but it works", then long life to Bonsai - give me more, please!
> And because Bonsai is written in OCaml, it becomes possible to use the same language and types on both the backend and frontend.
Finally! I was waiting for this to become possible!
Why wait? It’s been possible since forever. I wrote UI in OCaml eight years ago. https://github.com/kccqzy/drawing-guessing/blob/master/clien...
There's also Fable for F# but I believe this commenter is being sarcastic. Javascript is a common backend language.
A really great place to reach for this is Elixir/Phoenix now that the set theoretic type system is in place with LiveView. I have been doing some pretty large/complex codebases gradually working towards fully integrating working this way for the past 5 years and it has been great.
Similar attempts include Scalajs.
The general challenge becomes integrating the fractional front-end code written in your backend-language that compiles to JS with the rest of the JS ecosystem.
JaneStreet have a love of writing their own stuff from scratch so it doesn't apply to them but it might to you.
Hence most people end up with frontend-as-backend rather than backend-as-frontend.
KotlinJs and ClojureJs too. You always end up having to write wrappers or dealing with problems that would just work if you just used normal js/ts.
Clojurescript not ClojureJs, and Typescript doesn't run in the browser, it has the exact same issues/limitations: it must be compiled to JavaScript first.
Compiling TypeScript is just stripping out types
That is true only if you confine your TS code to be pure JS plus type annotations.
There are TS constructs that require compilation into equivalent JS, such as enums and namespaces.
Are there any other?
Namespaces are heavily discouraged. Some teams still use enigma but most people in the know us unions of strings.
I know the TS team regrets enums and I have not seen any plans for adding additional keywords that break type stripping.
The other one I read about was parameter properties in constructors.
And thank you for the correction! It was quite a few years ago that I used TypeScript, and I think the landscape was different then.
Making things even fuzzier in my mind, this was an Angular project, and it was hard to keep track of what was Angular and what was TypeScript.
It sounds like people these days actively avoid the parts of TypeScript that require translation into different JS code.
"fortunately" we can load swc compiled to wasm to deal with that.
absolutely nobody used those outside of a language demo. We also got --erasableSyntaxOnly last year to remove them from the language. The TS team considers them “mistakes“ for a really long time now
Thanks for the correction! It was a long time ago that I used TypeScript, and making things worse (for my memory) it was an Angular project.
So today, most people and teams avoid the parts of TS that require compilation into different JS, and --erasableSyntaxOnly lets you enforce this, right?
Yup, and Clojure/ClojureScript!
Couldn't WASM solve that problem once and for all? Is there some limitation that WASM has that JS doesn't?
Disclaimer: I am very inexperienced at front-end development.
> Is there some limitation that WASM has that JS doesn't?
You need a JS trampoline to call your WASM and make browser primitives available to it, and IIRC calls into browser code incur some extra overhead, but those are pretty manageable.
Additionally: for high level languages, source code is _much_ smaller than compiled binaries. If your initial needs are simple, your users are likely downloading more than 10x as much code.
WASM does not (yet) have access to the DOM or web APIs, meaning every direct interaction with the browser must first go through JS interop anyways, with a resulting performance penalty. There are proposals, and maybe we'll see it happen by 2035, but as of now WASM is best suited for heavier workloads where the edge in application performance outweighs the cost of JS interop with the browser rather than being a universal solution.
Such attempts are quite common. Some I remember:
https://ocsigen.org/ in Ocaml too
https://websharper.com/ for fsharp and csharp. Really good when I used in in fsharp
Maybe https://melange.re/v7.0.1/ too? (Not sure)
I took this as a joke in reference to the popularity of JavaScript on the backend for the last two decades.
that's because most people's intro to CS was how to build a webpage.
Modern web development is more like computer alchemy
> Finally! I was waiting for this to become possible!
Sarcasm?
Do you want to have a talk as to why JaneStreet wasn't built on JavaScript?
There's a great Signals and Threads (Jane Street's Podcast) episode about this work: https://signalsandthreads.com/building-a-ui-framework/
A framing from that podcast that helped me better understand the role of Bonsai is that it's really a framework for building incremental distributed state machines.
This framing is also in the post. From the readme:
> Bonsai itself -- this library -- is actually more generic than the above makes it sound. It allows you to build general-purpose incremental, composable state machines. Bonsai_web builds on top of that core library, specializing it for interactive browser-based UIs, but we also have Bonsai_term for building interactive terminal-based UIs.
The docs are a bit sparing but if you read the source its pretty cool to see how they did this.
Curious how this compares to Melange which is used by Ocaml shops as well to double up on Ocaml for both front and backend (ahrefs being the major user and sponsor). Does this mean giving up a lot of the JS ecosystem (React, graphql, etc)?
My understanding is that js_of_ocaml has been around longer and likely has a stronger ecosystem, but Melange was designed more specifically to play nice with the JS ecosystem.
I am sure it's very performant, but to me it's extremely ugly; Surely someone can fix margins and still have it be performant.
I’m no UI expert - what’s wrong with the margins?
inconsistent and maybe too tight, but I would think those are issues with the sample UIs and not this library
FWIW if you listen to Signals and Threads (the Jane Street podcast) they talk about this.
They have dedicated UX designers who do relatively little code but focus entirely on making the UX as optimal for the traders as possible and one of them was on an episode.
They talked about how most of the normal UX rules for SW go out the window because quants and traders have such different needs than "normal software".
This matters. Beyond performance (a universal good), the definition of "good" UX is strictly dependent on the user. Fintech platforms for traders who open a multi-window thick client application(s) and spend hours in it every day have little in common with consumer mobile apps designed to hold your attention for maybe a minute. It's mind-bogglingly lazy to choose something like MUI (Material UI) as if it were some kind of reasonable foundation irrespective of context.
Is there any professional software that wouldn't benefit from high information density? Most software design is optimized for the non-users who have to approve it.
You can have high information density and still not have words running into each-other. Taking all of the space out from around text and information graphics makes them less legible, even for experienced "power users".
I think the goal here is more an aesthetic of being for "serious traders only" rather than a UI optimized for speed and clarity. The UI designers probably have little choice but to match the preferences of the user group though, even when those preferences are empirically detrimental.
"non-users" are also big targets for software. If you run a software business, chances are you want people who've never used your software to start using your software. In that case, discoverability trumps high information density.
It only adds discoverability for the things that fit the screen. In other words, you have to make it simple. I.e. the lazy way to make something "user-friendly", do as little as possible and force the user to adapt to your ways instead.
this - most of the internal software at my job that's actually important is very dense, with keyboard navigation and shortcuts. The ones that somebody makes to sell themselves to upper management while angling for a promotion look very pretty, and never get used.
And yet, we make such beautiful and vacant software ourselves for our clients, because again, its b2b, so you aren't selling to the people who actually have to use the software. Im glad at least we arent forced that often to use specific internal tools, the better ones just get adopted organically
Fair point
I guess JS doesn't have to care about this at all because they're not selling their UI to others, just making it in house for in house use
From experience I can say that traders prefer extraordinarily little whitespace in their UIs.
A lot is said in this thread about user preferences for info density, but it is really a lot more than preference. there's also a big component of it that is user productivity focused. Info-dense UIs simply require a lot less scrolling around and pagination to look at relevant data. For example the ability to look at multiple charts side by side instead of having to scroll one chart after another into position enables users to spot differences that they wouldn't otherwise see.
I'd kill for tight UIs now. Everything insists on having 15000-mile wide margins, and at most two items visible on an 8K screen.
We are at a point when TUIs from 1990s running in 12" screens with at best a 480x240 resolution display more info than almost anything we have now.
This makes me think about the bizarre UI regressions in macOS in the last few years. I’ve got this giant, beautiful display and somehow it becomes less rather than more useful over time due to the software it’s presenting.
Margins on an info dashboard are wasted space.
Probably because they skew towards higher information density
What does JaneStreet do besides sponsoring nerdy Youtube channels and writing UI libraries?
Jane Street made it's way into today's Wall St Journal for its quantitative prowess and hiring of Math PhD's. What was stunning to hear was that their 2nd quarter profits were a record $10.3 billion, nearly double that of Goldman Sachs and Morgan Stanley (much larger firms).
They're a trading firm that hires really clever people to use math to trade. Their strategies are, presumably, very complicated and ever changing, and secrecy about them is the point - you get bigger margins for longer if people don't learn about your 'one weird trick' (that and also that there is presumably another side to your trades losing out on upside, and they might not be best pleased to learn about it.
> really clever people
Sam Bankman-Fried and his League of Legend rankings would like a word
They’re a market maker
Print money?
Governments print money, high frequency traders extract money. They would say that they provide liquidity to the market as a service.
hn: come for the tech news, stay for the pedantry
Interesting! What are Bonsai's dependencies? I use OCaml but - so far - none of Janestreets libraries etc.
Basically most of Jane Street's published code[0] via its dependency on Core.
[0]: https://github.com/janestreet/bonsai/blob/f31661450eb133fe89...
Can someone who understands web UI programming tell me if this would be good for my local agent to use to produce HTML based reports and outputs for me? Or for TUI outputs?
The main thing that Bonsai gets you is a nice programming model for making your UIs interactive. If you're mostly producing static reports (with maybe some basic interactive visualization widgets or something), it will not get you much over generating some variant on HTML + JS.
It makes more sense for applications where you have complex interactive logic in the UI, with a lot of different components that need to control and display some shared state.
There was this submission two months ago: https://news.ycombinator.com/item?id=48365904 . Basically they have a TUI frontend for the same thing. And the blog post says: "Bonsai_term would feel familiar to anyone who’d ever done web development here, and it had the huge advantage of being especially amenable to AI assistance. It was actually somewhat of a mystery to us how good the models were at writing Bonsai_term code, given how relatively obscure it is [...]"
Probably not, because your agent won’t have much info about this tool in its training set.
i maintain a codebase that uses Bonzai (https://github.com/hazelgrove/hazel/); agent based development works just fine, though probably a bit more token churn
Looks like it's Web-only, no mention of the native UI support (terminal UI excluded).
There's a full terminal implementation of Bonsai as well. I actually use it in my personal workflow these days to manage my contacts database! https://anil.recoil.org/notes/aoah-2025-9
Seconding the sibling comment - what are you using to run your personal site? It is very nice.
Unrelated but what do you use to manage your personal website? I love how everything is interconnected.
Custom infrastructure; I use my site as a playground for various things over the years. It ran as the first MirageOS unikernel for many years, and recently it's been running on a zero-allocation OxCaml webserver https://anil.recoil.org/notes/oxcaml-httpz to help me get to grips with the new extensions.
Not sure why, but reminds me of Fog Creek's Wasabi.
Oh it needs a userland trampoline!
> JSOO does not have tail call optimization
Jsoo supports self-recursive and mutually recursive tail calls, but for general tail calls it would need to trampoline every single call, since the only major browser with proper tail calls is Safari.
The “The thinking in bonsai” link 404s.
This says it is based on Elm. So it has the same clean immutable state structure?
I really want to like OCaml. One day I'll get around to learning it.
I wish there were more OCaml shops out there
This would of been cool back in 2014.
Ocsigen and Eliom were started 2005 and are pretty similar, although I can't find any well-known projects using Eliom :/.
Presumably this has learned some lessons from it, so it's probably better.
>"This would of been cool back in 2014."
Pretty much. UI libraries are more or less meaningless now, and I say this as someone who spent the last decade building them. Given a proper data API to work against, you're a single prompt away from a fully tested bespoke lib that's more maintainable and has no dependencies. The weeks of work you saved by reaching for one before is gone, and the tradeoff you had to make of dealing with upstream changes and lack of customization just isn't there anymore.
This is great. It focuses on utility and information density over design. It looks like someone took a terminal UI and transplanted it to the web, Bloomberg terminal style.
I'm pretty sure you can build tools with this that are fast and pleasant to use.
The "Why Bonsai?" I found really funny.
Let me re-write that section for you:
Why Bonsai?
At Jane Street we're super excited by Functional programming and by CAML in particular, so when we need low latency software, we use OCAML, when we need hardware, we write out own langauge - HardCAML, and when we need a Web UI, we build a Web UI framework in CAML. Because we fucking love CAML.
And there's nothing wrong with that. Many languages attempt to do the same thing. Why bother with 15 different languages if you can leverage one?
See also LiveView and Hologram for Elixir.
Agreed, looks like it's reinventing the wheels. Frameworks does not really matter today, AI agents write the code anyway.
But AI do benefit a lot from a strict compiler, and having a simple language improve a lot on type-safety, so I don't think it is to throw, even today.
But what about compilation speed? Hot reload? Maintainability? What is the bus factor for these kind of technologies?
Yea, its cool to have strict type safety, but what's the point of it doesn't have all the benefits from other Frameworks that took years to polish
JaneStreet has enough free cash to not worry about those issues. If one of their key persons dies in service, they can go out and contract the worlds leading expert in that domain, and their annualised rate isn’t even a rounding error on their bottom line.
ocaml has excellent compilation speed, and in my experience the code is very maintainable due to the module and type systems.
AI certainly writes Internet comments so that the flock does not have a single minute to get independent thoughts.
Better frameworks will still help AIs avoid silly mistakes. A smarter framework means you can be productive with a dumber/cheaper AI.
It looks like a nice little library; but oh boy must this be so limiting for the product teams that are forced to use. Everything looks like it's straight out of the 1990s.
Considering how ridiculously verbose their library is compared to React/Vue it's a good thing we have LLMs to deal with the boilerplate now.
All your money and retirement funds are safe; I’m too dumb to work at JaneStreet; every time I see OCAML I feel less than and confused.
Why does this library get posted here, what feels like every month? I remember it seeing at least twice before.
Been here 5 years, never saw it
HN is obsessed with anything done in niche programming languages, is my guess.
Also, Jane Street.
It might be a very performant UI framework in OCaml, but all the UI elements look extremely unpolished to me, like what a really smart high-school kid would build the GUI for his side project on Windows in 90s.
Thanks, I can use JS as a functional programming language.
Windows and especially Macintosh UIs from the late 90s/early 00s were by far the most functional UIs we've ever had since the dawn of the desktop era - user research was a serious discipline, accessibility was baked in to the frameworks, and the HID guidelines that devs adhered to gave us uniform conventions that we users could rely on.
If the biggest criticism of Bonsai is "it ain't pretty but it works", then long life to Bonsai - give me more, please!