Every few weeks Pi hits #1 here and I quietly grumble "mine does that too, but with lovely graphics.", so I'm saying it out loud: <a href="https://github.com/juggler-ai/juggler" rel="nofollow">https://github.com/juggler-ai/juggler
Like Pi, it's plugins all the way down, provider-agnostic, minimal system prompts, threaded sub-agents, code-mode, multi-client remote sessions, worktrees, a context window you can actually see and edit, etc etc
Where Pi is way ahead is the plugin ecosystem, and that takes people, which is hard to get amid the current deluge of agent action. So if there's any spare oxygen trailing off this thread, I'd love any Pi-heads who fancy a bit of GUI action to come and kick the tyres..
This project looks great, well done, BUT it seems very odd to be "grumbling" about Pi when you're explicitly targeting a separate audience (GUI users).
There are other GUI tools like Juggler that are quite popular, like cmux, that it seems better placed as an alternative to.
Thanks! My grumbling is really just that the sheer amount of noise and churn in this area at the moment is making it hard to get any attention, despite there being so many millions of potential users out there. Unlike many of the others, I'm just one bloke, not a VC-funded outfit with a marketing budget!
It looks like a good system, great if it works for you - but we now live in a period where I could build a similar tool to handle MY preferences in a weekend if I wanted to.
And, it would almost be EASIER than learning to use a new tool. I prefer PI because I didn't want batteries included in the terminal. Any decisions about what a user might want is a decision made on the user's behalf that dilute that core.
Juggler looks good - not personally for me. But that's OK, it's got 675 stars and a bunch of forks. That's attention, no?
I think people vastly overestimate how hard it is to get a tool right even in the age of AI. Yes, it's very easy to get a throwaway tool done in a session. For anything above trivial complexity (like a coding harness) you will immediately notice warts with it.
It's true: with proper planning, thought, and focus, you might be able to generate enough code in one weekend to create a tool that is at parity with something like Pi - but chances are the amount of focus, planning, and thought is going to be greater than one weekend worth. And I don't know about you - but I don't really want to spend my weekend building a coding harness, I have other ideas and projects I'd rather execute.
Yes indeed. In the case of juggler, if you check my CV you'll see that I'm very far from being a vibe coder - I've been shipping hugely complex products for 30 years, I've written frameworks, entire UI libraries, audio libraries, a DAW, a couple of compilers.. But this took me about a year to get right. I must have re-architected the whole thing more than 10 times, it took every bit of my experience to not screw up the data model. It's actually one of the most difficult and interesting projects I've done. So yes, if you can get claude to churn out a juggler-like app in a weekend, it probably ripped off the source code!
I think you probably meant to write “underestimate”. But I think you are overestimating how difficult it is!
I’m not exaggerating when I say all I had to do was prompt “build me an LLM coding harness like Claude code, but make it so that I can resume sessions from OpenCode and Pi as well” and that’s it, and it was done within an hour.
Didn’t have warts? Sure, a couple of really minor things. One of the things I realized quickly was that it didn’t have a WebFetch tool built in, so I prompted “add a webfetch tool” and it did it within 5 minutes. Warts are trivially fixed.
The things that really differentiate software now are the ideas behind them, not the implementation. So if (like the OP) you’re looking at a tool and going “well mine does that too…”, so what? If it’s just a re-implementation of the same ideas, anyone can recreate it (and add on their own customizations to boot).
Personally, I am getting frustrated with the amount of people trying to pitch me their tool they made. Tools are a dime a dozen. They’re cattle now, not pets. Share with me your ideas, and I’ll share with you mine, but I don’t care about your tool.
> Did it have warts? It’s been working nearly flawlessly, but there have been a couple of really minor things. One of the things I realized quickly was that it didn’t have a WebFetch tool built in, so I prompted “add a webfetch tool” and it did it within 5 minutes. Warts are trivially fixed.
Sure, and then tomorrow it's compaction, then worktrees, then an annoying buffer scrolling bug, image clipboard handling, etc etc. I don't buy for a minute that a one-liner prompt creates a perfect program because I know from experience it doesn't. You are inheriting the maintenance of a non-trivial program. That's perfectly fine if you want to tinker and spend time on that but I would rather offload that to someone who wants to spend more time thinking about those problems than I do.
> Personally, I am getting frustrated with the amount of people trying to pitch me their tool they made. Tools are a dime a dozen. Share with me your ideas, and I’ll share with you mine, but I’m not gonna adopt your tool.
I'm not sure I follow this line of thinking. Obviously, juggler is the result of an author who has spent a great deal amount of time thinking about an idea. It is the manifestation of the author's ideas. I agree that the amount of vibe coded slop out there that people think is marketable is ridiculous, but Juggler doesn't really look like that. It's still worthwhile to evaluate well thought out tools IMO.
> Sure, and then tomorrow it's compaction, then worktrees, then an annoying buffer scrolling bug, image clipboard handling, etc etc. I don't buy for a minute that a one-liner prompt creates a perfect program because I know from experience it doesn't.
Every thing you listed works out of the box. Believe it or not, it’s true! And creating tools like this is only going to get easier as LLM models progress.
Your comment very much reminds me of comments from 2023 saying that LLMs would never be able to write working code, or be able to generate lifelike pictures, or be able to find security issues… but look around you, they’re doing all of those things already!
I'm using these tools everyday just like you and everyone else and I 100% know they don't produce code that works perfectly out of the box. That's horseshit.
To answer your question about what I was trying to do that was different:
- a multi-client, state-machine based architecture rather than a naive LLM loop
- a bunch of UX features that I wasn't seeing in other harnesses (although of course they're all churning away in terms of UX so this changes all the time)
The difficult bits were:
- getting the data model right. This was a lot harder than it looks!
- getting an automated UI test framework in place that was robust enough to trust that incoming changes wouldn't break something. Originally I hoped to get away without this (I've never worked on a project before where we had automated UI tests!) but it was just a shitshow. I spent a couple of months getting this right, but it turned the whole project around and it's now incredibly quick and safe to make changes.
julesrms · · focus · HN ↗
Like Pi, it's plugins all the way down, provider-agnostic, minimal system prompts, threaded sub-agents, code-mode, multi-client remote sessions, worktrees, a context window you can actually see and edit, etc etc
Where Pi is way ahead is the plugin ecosystem, and that takes people, which is hard to get amid the current deluge of agent action. So if there's any spare oxygen trailing off this thread, I'd love any Pi-heads who fancy a bit of GUI action to come and kick the tyres..
lucideer · · focus · HN ↗
This project looks great, well done, BUT it seems very odd to be "grumbling" about Pi when you're explicitly targeting a separate audience (GUI users).
There are other GUI tools like Juggler that are quite popular, like cmux, that it seems better placed as an alternative to.
julesrms · · focus · HN ↗
NichoPaolucci · · focus · HN ↗
It looks like a good system, great if it works for you - but we now live in a period where I could build a similar tool to handle MY preferences in a weekend if I wanted to.
And, it would almost be EASIER than learning to use a new tool. I prefer PI because I didn't want batteries included in the terminal. Any decisions about what a user might want is a decision made on the user's behalf that dilute that core.
Juggler looks good - not personally for me. But that's OK, it's got 675 stars and a bunch of forks. That's attention, no?
boredtofears · · focus · HN ↗
It's true: with proper planning, thought, and focus, you might be able to generate enough code in one weekend to create a tool that is at parity with something like Pi - but chances are the amount of focus, planning, and thought is going to be greater than one weekend worth. And I don't know about you - but I don't really want to spend my weekend building a coding harness, I have other ideas and projects I'd rather execute.
julesrms · · focus · HN ↗
cobolcomesback · · focus · HN ↗
I’m not exaggerating when I say all I had to do was prompt “build me an LLM coding harness like Claude code, but make it so that I can resume sessions from OpenCode and Pi as well” and that’s it, and it was done within an hour.
Didn’t have warts? Sure, a couple of really minor things. One of the things I realized quickly was that it didn’t have a WebFetch tool built in, so I prompted “add a webfetch tool” and it did it within 5 minutes. Warts are trivially fixed.
The things that really differentiate software now are the ideas behind them, not the implementation. So if (like the OP) you’re looking at a tool and going “well mine does that too…”, so what? If it’s just a re-implementation of the same ideas, anyone can recreate it (and add on their own customizations to boot).
Personally, I am getting frustrated with the amount of people trying to pitch me their tool they made. Tools are a dime a dozen. They’re cattle now, not pets. Share with me your ideas, and I’ll share with you mine, but I don’t care about your tool.
boredtofears · · focus · HN ↗
Sure, and then tomorrow it's compaction, then worktrees, then an annoying buffer scrolling bug, image clipboard handling, etc etc. I don't buy for a minute that a one-liner prompt creates a perfect program because I know from experience it doesn't. You are inheriting the maintenance of a non-trivial program. That's perfectly fine if you want to tinker and spend time on that but I would rather offload that to someone who wants to spend more time thinking about those problems than I do.
> Personally, I am getting frustrated with the amount of people trying to pitch me their tool they made. Tools are a dime a dozen. Share with me your ideas, and I’ll share with you mine, but I’m not gonna adopt your tool.
I'm not sure I follow this line of thinking. Obviously, juggler is the result of an author who has spent a great deal amount of time thinking about an idea. It is the manifestation of the author's ideas. I agree that the amount of vibe coded slop out there that people think is marketable is ridiculous, but Juggler doesn't really look like that. It's still worthwhile to evaluate well thought out tools IMO.
cobolcomesback · · focus · HN ↗
Every thing you listed works out of the box. Believe it or not, it’s true! And creating tools like this is only going to get easier as LLM models progress.
Your comment very much reminds me of comments from 2023 saying that LLMs would never be able to write working code, or be able to generate lifelike pictures, or be able to find security issues… but look around you, they’re doing all of those things already!
boredtofears · · focus · HN ↗
julesrms · · focus · HN ↗
- a multi-client, state-machine based architecture rather than a naive LLM loop - a bunch of UX features that I wasn't seeing in other harnesses (although of course they're all churning away in terms of UX so this changes all the time)
The difficult bits were: - getting the data model right. This was a lot harder than it looks! - getting an automated UI test framework in place that was robust enough to trust that incoming changes wouldn't break something. Originally I hoped to get away without this (I've never worked on a project before where we had automated UI tests!) but it was just a shitshow. I spent a couple of months getting this right, but it turned the whole project around and it's now incredibly quick and safe to make changes.