Loopy Pro: Create music, your way.
What is Loopy Pro? — Loopy Pro is a powerful, flexible, and intuitive live looper, sampler, clip launcher and DAW for iPhone and iPad. At its core, it allows you to record and layer sounds in real-time to create complex musical arrangements. But it doesn’t stop there—Loopy Pro offers advanced tools to customize your workflow, build dynamic performance setups, and create a seamless connection between instruments, effects, and external gear.
Use it for live looping, sequencing, arranging, mixing, and much more. Whether you're a live performer, a producer, or just experimenting with sound, Loopy Pro helps you take control of your creative process.
Download on the App StoreLoopy Pro is your all-in-one musical toolkit. Try it for free today.
Comments
Very interesting question. I’ve no idea but it’s the first one I’ve come across here which states that it’s 100% vibe coded. Others will know better than me no doubt.
Let me make a bold claim: GRE is the first 100% vibe‑coded plugin - for better or worse - unless someone can prove me wrong
So @Augmatic you essentially had the entire architecture planned out extensively and fed it all to Claude in one initial go? The issue I’m running into with vibe coding is getting a bit lost trying to review what ChatGPT is doing code wise. I’m kind of going feature by feature on my app and not sure if I should go try to clean up mid development to save myself a potential headache in the future or trust that things are working under the hood and press on until finished.
Far from that! I had no idea what to expect from Claude or how far it would take me. My first plan was just to wrap Grids into a VST format, for my own use. That worked quite quickly, so I tried to create six independent channels, then added a Clock Divider, then a Velocity Bender, then Linear Drumming, and so on - feature by feature.
I was doing a lot of architectural research from one step to the next. I got stuck hard twice: first on Swing, which Claude kept trying to implement after (!) Grids engine, over and over again, despite my attempts to explain that it needs to slow down and accelerate the Clock.
Second, when I tried to create an acceptable GUI. The first version was basic JUCE and it was ugly. I thought I’d go for a webview, and I even created a full prototype (which is now partially visible on the website https://augmaticaudio.github.io/gre/). I did a lot of research on GUI options for JUCE and ended up giving up on webview, because Apple doesn’t like it on iPad/iPhone, and AUv3 was my main target. VST wasn’t really the goal - I used it only for testing functionality on Mac, because it was faster than building in Xcode each time. But when I finally started creating a nicer GUI in JUCE, Claude got it right quite fast. I regret not trying that from the beginning; it would have saved me a lot of time.
I created and tested over 500 builds across six months. Sometimes, the changes were quite small, this was driving the number of iterations. I have a job, so I was doing this in the evenings and on weekends, whenever I wasn’t too tired to sit in front of a screen. Fortunately, winter was really bad this year ;-)
I wasn’t checking the code at all, since I have zero C++ knowledge, and trying to learn it along the way would have taken too much time. I asked Claude to check and clean up the code a few times as I moved from one feature to the next. During troubleshooting, Claude created a lot of rubbish that had to be removed several times. I’m sure the code is far from optimal, but... it doesn’t crash :-) Apple approved the iPad app in the first round. The Mac version had to be updated once - Claude enabled Bluetooth for some reason, and since it wasn’t actually used, Apple told me to remove it.
Many architectural decisions made by Claude along the way were wrong; I can tell based on my previous Reaktor and Max/MSP experience. Here you can see the high-level MIDI signal flow: https://augmaticaudio.github.io/gre/user-manual/index.html#signal-flow I was asking Claude a few times to create Mermaid diagrams to illustrate the architecture, to show all modules, to make sure it's on track. Here is my biggest problem with AI: it will never say "I don't know" and always tries to "fix" something, eventually breaking cardinal rules from claude.md file and foundations of the architecture. This happend a few times when implementing something as basic as Swing, as I mentioned. I suppose there is not enough DSP/C++ open source available to train AI to the same level of expertise as it's trained for more common applications, like desktop apps or webpages. I would prefer if Claude would say "I don't have a good idea how to proceed, let's do some research", instead of trying all sorts of ideas, including bad ones, wasting time and breakting stuff that works and has been extensively tested.
Are you confident in being able to maintain this app over time? Apple is notorious for introducing breaking changes with their updates. What is your plan if vibe-coding a fix won't solve it; get to know your own code before that happens, or hopefully deal with it when it happens.
I hope to earn enough to eventually hire a developer if Claude can’t handle it :-)
That’s the main reason the app is paid - so I can keep it alive.
If I would release if for free, and people would make it part of their creative workflow, and then something would happen that I won't be able fix on my own - this would probably be much more upsetting to those users than spending a few bucks.
Update: I can confirm Claude did a vastly better job of my midi plugin idea. It basically did 90% of this while I was out doing the shopping. Had to re-steer it a couple of times when I got back, but pretty much nailed it.
What do you mean by doing it out shopping? Did you initialize it and let it run? Or prompt remotely and test back at home base?
The initial version took two prompts. The first in regular Claude chat where I told it what I was planning to do and what it needed to go off and research, and produce a markdown doc out the back of that. Then in Claude Code I gave it a list of requirements and the doc from step one, and switched it into YOLO mode. I set it off and went out to do the weekly shop, came back to a 90% complete plugin.
The final 10% was a bit of back and forth as I tested out the functions and reported bugs. The end result is something that seems to generate considerably better basslines than anything I've tried previously (though this is very much tailored to the kind of music I make). Now I just need to work out how to get it onto my iPad.
I can relate with this. As I’m getting to the GUI part of the app in particular I’m running a new build every time I start lining things up, changing some colors, new buttons, etc. It has been really fun focusing primarily on design instead of coding up the logic for the design. If ChatGPT has trouble a lot of times I can annotate a screenshot of the app and it will fix it as needed. To be honest, I’m not familiar with JUCE. I’d mocked up a color scheme and layout with a pretty specific vision starting out and just let ChatGPT implement it. I’m not sure if you’d recommend a way to implement JUCE, or if it would really bring much to my app since I’m going for a really utilitarian design with everything mostly static.
I have noticed the code gets bloated troubleshooting. There’s been a few errors that started to feel a bit recursive and went beyond clicking the red “x” and “generate fix”. I needed to talk to the AI and have it figure out a fix - I still didn’t code anything up, but I had it talk through the code until the error was pinned down.
The AI removes me of an over zealous intern. They have a million ideas they suggest to implement, do things a bit shoddily, think they’re hot shit, fumble pretty often, but can produce something meaningful with proper guidance.
I need to pick this up once finances are straight so I can really see what the app is capable of.
I’ve mostly been using ChatGPT in Xcode for my projects, but it seems like Claude is the more popular option.
I found this while going down a rabbit hole last night. It’s repomix, which lets you feed it a GitHub repo that “packages your entire codebase into a single AI-friendly file.” So you can give it a repo that it consolidates into one file you can then give Claude, ChatGPT, etc that makes it easier for the agent to analyze the repo.
https://repomix.com/
My way of avoiding code bloating due to multiple attempts by Claude to do something was following: I would ask Claude to create a document with details of a wrong response to my prompt, then I would roll back from git and start over, this time asking Claude to NOT repeat the same mistake from the document.
Sometimes it took several bad iterations, and the documentation of bad solutions was getting pretty large. After finishing a feature, I was asking Claude to create a summary document based. The most important "lessons learned" will be captured in a concise format, some bad solutions were not relevant anymore and could be removed.
To me, those documents are Claude's "experience" with my project which I can recall when touching a feature again. Think of it as CLAUDE.md file per feature. There is a cost of recalling this experience - higher utilisation of the context, frequent compacting - but avoiding the same mistakes or breaking a fundamental solution is priceless.
Here is the answer, I couldn't resist
https://www.youtube.com/shorts/pAWw6Ao5tr4
I'm very interested in this specific statement: "though this is very much tailored to the kind of music I make"
What sort of instructions did you use to describe that? If I'm ever going to take the time to build something for myself with an LLM, this is the part that would be most important to me.
I make a lot of acid house so I wasn’t so much describing what I wanted musically, but the rules for writing the bassline (whether for a 303 or something else) which should be much closer to the rules of generating a catchy melody / hook - stepwise motion, gap fill etc. So hitting generate creates a new pattern, but every pattern tries to follow some of the rules we set up.
For anyone else interested in getting a robot to build plugins for them, there's a ton of good info on guiding the robot here: https://simonwillison.net/guides/agentic-engineering-patterns/
I'm not surprised to learn that this app came about in this way. It doesn't have the kinds of features I would expect to see from an established and experienced developer. It feels raw and unbaked in some regards, the sort of thing that an experienced and established developer would never release. However, I also appreciate it for what it does and it reflects thought into what makes such an app useful.
I'm not particularly offended by it, and I paid for the app. Being against AI-assisted coding is a waste of time. We can't choose to live in a world where this tech doesn't exist.
If someone can earn some money doing it, good for them. This is the phantasm AI bros insist will prevent the vast bulk of humanity from falling into poverty: Lost your job? Just vibe code the next great app. Nevermind that capital controls all the entry points to this infinite well of income, and will inevitably charge higher and higher rents for access, or else just come over the top of all the mom-and-pop vibe coders with $1 billion of capital investment.
We live in hell, and this app isn't responsible for that.
It inevitably means that we'll pay more for pro apps that reflect a greater investment in labor hours. I'm surprised we've gone this long enjoying such low prices on iOS.
very interesting. do you have an example of how you described those rules? curious as to whether you leveraged music theory or terms to do it, or used more pseudocode type language.
I completely disagree with those views about the app. I don’t see it as half baked, and to me it has all nearly the features it needs right out of the gate. When I read that it was vibe coded I had very low expectations. But I got it out of curiosity and had my views completely changed.
I’ll never vibe code myself, but if stuff this good can be produced already by people that wouldn’t have done it otherwise, I’m all for that.
Same. This is a FAR more competent release than many apps I've reviewed fully coded by human devs, it's not half baked at all imo. But...i do agree that AI will, in our flawed society, end up having more cons than pros, and we'd probably be better off if these LLMs had never been invented.
-
Regardless of anyone’s individual feelings about vibe coding I think most of us can agree it was pretty brave of @Augmatic to be honest and up front about using it to help create the app. Cheers!
Still, with all the flaws in our society, don't underestimate the impact the LLMs and deep learning neural nets will have on scientific discovery. Think drug discovery, material design, energy utilization, dna and protein understanding, ...
Not just that, but to come here and provide insights into the process, the thinking, etc. Much appreciated. I don't need the app but I bought it just to support the dev and the continued insights into how software (and importantly, the market for it) will evolve over time.
I'm more or less under a mandate to adopt AI at work, and so I've been experimenting with Claude in my free time. My background is in systems and network administration, so I have a lot of depth in SDLC, complex architectures, data structures, etc, but not a lot in lower-level programming languages. I'm having some success building an iOS app and it's solving some complex problems (with quite a bit of iteration), but it has struck me several times that if I didn't have that underlying systems development experience the AI would have steered me in to a brick wall a few times along the way. So for now at least, it's a force multiplier for experienced developers rather than a means to take developers out of the picture. We'll see what the future holds.
It's still worth noting that, as far as I know, none of these LLM companies have a realistic path to profitability that doesn't rely on miracles occurring in the areas of compute power per watt and/or energy generation. So whatever we'll contend with in the future will be whatever emerges from the ashes once that bubble pops. It could end up that ultimately "A great eye, lidless, wreathed in flame" or whatever the current vision is, will just not be fiscally viable.
If the dot.com "crash" is any indicator, major corrections are inevitable, but no emergent tech really dies or even substantially slows down.
I started testing Claude Coding yesterday after solely using ChatGPT for about a year now and I’m amazed. Working in a terminal and letting it have at it is a sight to behold.
Yes, the underlying tech is here to stay. But the current business models may or may not survive. It could be that what ends up being sustainable is something less than covering the land with gigantic datacenters.
Actually, none of the AI companies has a real business model. I bet that even premium Claude subscriptions are not enough to power their data centers for an hour.
All AI companies burn cash (and the planet) like crazy. Where is the money coming from? From people who believe that scaling the models will lead to AGI, and that AGI will “cure cancer during breakfast” (or maybe "take over stock exchange in a week").
Indeed, simply adding more CPU, RAM, and training time allowed models to jump to entirely new levels — think GPT‑2 → GPT‑3 → GPT‑4. However, GPT‑5 didn’t make that kind of leap anymore. More people are realizing that building more data centers won’t lead to AGI; it will just increase energy prices.
So Sam Altman will introduce “adult erotica” (what could that be? hmmm) and advertisements. Sure, adult content might bring in some revenue, but still not enough to power the data centers for a day, let alone pay back the money they are borrowing — with profit.
So, use AI while you still can and a naive billionaires are paying for it!
Yeah, that’s kind of my take. Anyone who’s investing because they think AGI is just going to “emerge” from a LLM is getting played. The rest of us can have a bit of fun on their nickel in the mean time. Also maybe it will crash the economy. But y’know, YOLO