The Endless AI guitar pedal has potential

I’m not sure anyone was really asking for an AI guitar pedal. But it was inevitable that someone would build one. One of the first to take the plunge is Polyend, a well-respected music gear maker with a reputation for building niche, idiosyncratic devices. The company has built grooveboxes around old-school trackers and a multi-effect pedal that you can step sequence. So there was at least some hope that if anyone could do an AI effect pedal right, it would be Polyend.

Polyend’s Endless is a $299 programmable guitar pedal running an ARM processor. It’s paired with Playground, a number of interconnected AI agents that turn any text prompt into a functioning guitar effect. If you have an idea, you don’t have to hope that someone’s already built that pedal; you can simply prompt it. Maybe there’s a specific combination of effects that you’ve always wanted, but no company sells it because there’s no demand for a combination ring modulator / auto-wah. I’m not convinced this is what guitarists are yearning for, but it’s a well-intentioned first attempt to marry an effect pedal to an LLM.

To be clear, the AI isn’t actually in the pedal. Instead, Polyend trained a custom LLM to code effects you can then load onto the pedal. You can also build effects yourself in C++, but most people will either download existing free Plates (as Polyend calls the effects) from the community site or prompt them in Playground. You can also pay $20 for a physical faceplate to pair with a downloaded effect.

Right now, the Plates gallery is home to about 60 effects, mostly developed by Polyend. They cover everything from simple saturators to tape loop simulators and guitar synths. There are even self-playing drum machines. Some of my favorites include Grunt (a lo-fi octave down effect), the Infinite Hall reverb, and Tessera (granular pitch-shifting reverb). There’s also Stardust, an enormous-sounding granular delay, reverb, and tremolo combination that would be hard to find elsewhere.

Polyend is opening up the gallery to third-party contributions as well, so you could whip up an effect in Playground and submit it for consideration.

Playground is the reason most people will be eyeing Endless. It’s a web frontend for several AI agents working in tandem, trained on Polyend’s own effects library. The various AI components interpret prompts, select effect algorithms, generate code based on those building blocks, and then validate that code to ensure it will run properly without blowing out your eardrums.

If you’ve ever used a chatbot before, the Playground web app should seem familiar enough. You describe the effect you want and its controls (you have three knobs, plus short and long presses on the footswitch to work with), and it will come back with a couple of options for turning your idea into something functional.

Generally, you get three options. You can just pick one and let Playground do its thing. But you can also make tweaks at this ideas stage, before it starts generating any code and, more importantly, costing you money.

Generating effects costs tokens. The pedal comes with 2,000 tokens, and you can buy more at $20 per 2,000. That should be enough for a few effects. The system is a bit opaque, but the more complex an effect, and the more iterations it takes to get it right, the more tokens it costs. A simple fuzz might only cost you 20 tokens, but a granular looper with rhythmically synced glitches might cost you 500.

What is likely to chew through your tokens isn’t the original generation, it’s the iterating to get what you want. Especially if what you’re looking for is balanced weirdness. Trying to get Polyend’s Playground to understand the proper resonance levels for a bandpass filtered delay, or how much ring modulation you want in your evil fuzz, can be frustrating.

Polyend provided me with 10,000 tokens ($100 worth) to evaluate Playground as part of this review. In total, I went through a little over 3,500 tokens and got three effects that I enjoyed and a handful of duds. If I were paying for tokens out of pocket, I might have been less inclined to keep iterating. Also, Playground can be slow to spit out code. Depending on the complexity of an effect, it could take anywhere from five to a little over 10 minutes each time. Often, I’d get bored and give up after five or six attempts if I still wasn’t getting what I was looking for.

The effect that posed the most trouble was what the chatbot called “Resonant Taps.” I prompted it for “a clean multi-tap digital delay where the delay lines have narrow, highly resonant bandpass filters with subtle modulation.” The first version gave no control over the feedback of the delay. And when combined with resonant filters, it was almost always on the verge of self-oscillation. Incredibly unpleasant. Swapping the resonance control for a feedback knob didn’t help matters much.

Eventually, after six generations, I tried prompting it with a description of the sound I was after, rather than the specific effect combination. I wanted “gently modulating resonant pings that follow the pitch of my guitar, almost like a string section.” Instead, I got what sounds like a ’70s synth having a nervous breakdown. It was fun, but not really what I was looking for. So after spending about $7 worth of tokens, I put that idea aside for something more immediately feasible, like a fuzzy, broken CD-skipping effect.

Even that took some effort to coax to life. My original prompt for “random stuttering and glitching” with “tap tempo” and a blend knob for a ring modulator was underwhelming. Even at its slowest, the stutters were machine gun-like and never actually felt rhythmic.

It took six generations and specifically asking for the stutters to be locked to 16th, 8th, or quarter notes for me to get what I was looking for. And despite asking to increase the ring mod level in each generation, it still feels a touch too subtle. That said, I do quite like the end result. It’s delightfully chaotic and just dirty-sounding enough.

There are also some quirks worth calling out. Endless can only load one effect at a time. Loading new ones is simple enough — you just connect it via USB to your computer, it shows up as an external drive, and you drag files over. But, even though the pedal automatically reboots after you load a new effect, most don’t seem to work properly until you manually power cycle it by unplugging it. Similarly, you might need to manually eject Endless and then reconnect it to your computer before loading a new effect. This dramatically slows down the process of iterating and testing.

Obviously, the sounds I’m trying to craft are complex — there are a lot of variables. But that’s kind of the point. There are a million and one digital delay and fuzz pedals out there that will do those things better than the Endless ever could. If you just want standard-issue effects, you’d be much better off building a small collection of even cheap dedicated pedals than relying on AI. The sole reason to get Endless is to try and create the effect of your dreams that doesn’t exist yet.

It’s important to keep your expectations in check, though. The effects Polyend’s Playground whips up based on your prompting are never going to compete with the careful programming of professional audio developers, or even dedicated hobbyists. What Endless is best for is quickly experimenting with ideas that you might develop further, either by programming it directly in C++ for the Endless, or on another platform like Max MSP or Pure Data. It’s just a shame Polyend won’t expose the raw code of effects created in Playground for further tweaking.

You might be better off exploring some of these ideas on a modular guitar pedal like the Poly Effects Beebo, the Empress Effects ZOIA, or even the Eventide H90, which let you assemble building blocks to create effects patches. Now, those devices all cost quite a bit more than Endless, at $449, $549, and $899, respectively, but they also have thriving communities and stable firmware, and produce top-tier effects. I’m far more likely to spend an hour manually building an effect on my ZOIA than fighting with an LLM.

And lastly, understandably, the AI aspect might turn some people off. But to its credit, Polyend seems to be making a good-faith effort to use AI as ethically as possible. In addition to building code solely from effects it developed in-house or from open-source components, the company is also trying to offset its environmental impact. Some elements of Playground are executed using external AI agents via API, and there’s little to be done about that. But founder Piotr Raczyński says that the Playground servers are on site, not in some third-party data center; Polyend says the servers are “almost 100% energy self-sufficient thanks to solar power and heat pumps.” It’s basically the “free-range, pasture-raised” of LLMs.

If you’re hardcore anti-AI, none of this will change how you feel about Endless. It’s probably part of the reason Polyend avoided using the phrase “AI” at all when it was announced. But if you’re a little more AI agnostic, or have fully embraced the world of LLMs, this might ease any guilt you feel. The primary advantage of the Polyend Endless is that it’s a cheap and approachable way to create custom effects. If a large community sprouts up around it, as one has for the ZOIA, it could be a great source of experimental one-off effects. But considering the music world’s general distaste for AI, that’s a big “if.”