The guitar world has spent the last few years treating capture tech like a holy war. One camp wants modelling. One camp wants captures. One camp wants to shout AI a few more times and hope that somehow counts as innovation. It is all getting a bit silly.
That is why Helix Stadium has my attention right now. Not because Proxy is already out in the wild and flattening everything in its path. It is not. As of March 7, 2026, the latest public Helix Stadium firmware notes Line 6 has posted are still 1.2.1, and Proxy is still preview territory. The Stadium Floor itself only started shipping in mid-February, so plenty of people are still in the honeymoon phase with the hardware before any of the capture stuff even enters the picture. But the strategy around Proxy looks smarter than the usual race to see who can scream capture the loudest.
What Line 6 seem to be doing with Stadium is building two lanes at once. On one side, you have Agoura amp modelling, which is their new deep-dive modelling approach. On the other, you have Proxy, which Line 6 are pitching as cloud-based cloning for amps, cabs, and effects. That split matters because the best modern gear is not really choosing between modelling and capture anymore. It is quietly mashing both together and pretending it was always obvious.
Agoura is the part I think people should take seriously, even if Proxy is getting the hype. Line 6 are not describing it like a small refresh to the old Helix formula. Their own material says Agoura looks at things like dynamic speaker impedance, the output impedance of every gain stage, tube current flow at different voltages, and the interaction between the power amp and power supply for sag and ripple. That is not just chasing the EQ curve of an amp. That is trying to model the ugly, complicated behaviour underneath the note, which is usually where digital gear gets caught out.
There is also some proper hardware grunt behind it. Stadium is running a new 64-bit floating-point architecture with a dedicated GPU, a machine-learning accelerator, and FPGAs, which is a very different proposition from the old Helix era. On launch, Agoura brought 16 new guitar amps and 6 bass amps to the platform, with firmware 1.2 adding another 7 Agoura amp channels from three new amps (Super Reverb, Mark IIC, and Silver Jubilee), bringing the Agoura channel count to 50 alongside the 111 legacy HX channels. That matters because it tells you Line 6 are not just bolting a capture feature onto an unchanged box. They are trying to rebuild the core amp side of the product as well.
This is also why I think the usual modelling versus capture argument is a bit dated now. Captures are great at getting you to a sound quickly. Modelling is great when you want a more interactive virtual amp that keeps making sense as you move the controls around. Real guitarists want both. Sometimes I want to build a sound from scratch and tweak it like a nerd. Sometimes I already know the exact sound I want, and I would rather not spend my evening rebuilding it block by block like I am assembling IKEA furniture with a plectrum.
That is the lane Proxy seems designed for. If it works the way Line 6 are suggesting, the obvious use cases are not abstract amp concepts. They are finished rigs. Your favourite amp through a reactive load. A cab and mic chain you already trust. A preamp pedal you know like the back of your hand. A drive pedal that just does the thing. That is where capture tech is at its most useful, because the goal is not to recreate every theoretical state of a circuit from zero. The goal is to nail the real thing you actually use.
What we know publicly about Proxy is still limited but the details that are out there paint a clear enough picture. It is cloud-based: the user does the capture locally on the Stadium, uploads the data, the cloud processes and trains the clone, and it syncs back to the device when it is done. You can keep playing while the cloud does its work, which is a practical detail that matters more than it sounds. Klein has confirmed there are no subscriptions involved, which removes one of the obvious concerns people had early on. He has also said Proxy is tailored specifically to Stadium hardware rather than being a generic approach, which gives Line 6 an advantage over something like NAM where the model has to work across arbitrary playback environments. NAM compatibility is still a maybe, but Line 6 want to let Proxy run its course first.
The most useful public clue so far came from Eric Klein, who said Stadium 1.3 only introduces the first phase of Proxy. His description of phase one is refreshingly practical: amp plus cab plus mic chains, amps with load boxes, preamps or preamp pedals, and distortions. He also described Proxy as using a mix of phase or frequency-domain and time-domain methods, which is the sort of nerdy detail I actually like hearing because it suggests Line 6 are not pretending there is one magic algorithm that solves every problem equally well.
That technical split is worth slowing down for a second. Phase and frequency-domain analysis is exactly where you would expect a system to get very good at the static fingerprint of a rig: EQ shape, resonances, mic coloration, cab behaviour, and all of the other spectral stuff that makes one chain sound different from another. Time-domain methods are where you chase the moving parts: pick attack, compression, transient response, bloom, and the way a sound shifts as you lean into it. If Proxy is genuinely combining both, that is a more believable approach than pretending everything important in a guitar rig lives in one mathematical bucket.
Klein has also mentioned some practical details around how clones are organized within presets: saved clones sorted into relevant model categories, clone blocks that can sit alongside regular Helix blocks in the signal chain. If those details hold up in the shipping version, it sounds like Proxy clones will be treated as first-class citizens in the preset ecosystem rather than being walled off in some separate capture ghetto. That matters for the hybrid use case I will get to in a minute.
The rest of the market is heading in this exact direction, which is why I think Line 6 may have timed this well. Neural DSP already moved Neural Capture V2 to Cortex Cloud and the new process is a meaningful technical step forward, not just a version number bump. V2 captures are trained via the cloud using a more complex algorithm than V1, and Neural now require you to fill in mandatory metadata before starting a capture, including the device type. The system actually uses that metadata to adjust its processing, which is a smart move because it means the algorithm is not trying to solve every piece of gear the same way. They shipped 669 V2 captures across 41 devices as a proof of concept library spanning 17 amps, 4 compressors, 11 fuzz pedals, and 7 overdrives. The new capture categories now include dedicated Overdrive, Fuzz, and Compressor types that did not exist in V1, which tells you exactly where Neural felt the original system was weakest. V2 blocks also use noticeably more CPU on device, which is why V1 still exists as the lightweight option for straightforward amps and overdrives where the extra resolution is not worth the DSP cost.
That is a pretty loud admission, in the best possible way. It tells you cloud processing helps, and it tells you everyone is still chasing the hard part, which is not just the tone but the feel.
That last bit matters more than marketing departments want to admit. It is easy enough to make a static clip sound impressive. It is much harder to make something react properly when you dig in harder, ease off with your picking hand, or roll the guitar volume back to six and expect the whole sound to clean up without turning into soup. Anyone can wow you for ten seconds on YouTube. The real test happens when the guitar is hanging off your shoulder and your right hand starts making demands.
Kemper are telling the same story from a different angle, and they just made their biggest move in years. Profiling 2.0 hit public beta on March 4 via Profiler OS 14.0 for MK2 units and the Player. The headline number is that the new profiling engine analyzes over 100,000 individual frequency points per profile, which is a massive increase in resolution over the original system. The refining phase is gone entirely, which is notable because that was always a bit of a crutch that hinted the initial capture was not complete enough on its own. They have also added adjustable Resonance Frequency and Resonance Intensity parameters for the cabinet section, which gives you direct control over the low-end thump and character of the cab rather than leaving it baked in and hoping for the best.
One detail that separates Kemper from the rest of the pack: Profiling 2.0 offloads part of the model calculation to your Mac or PC via Rig Manager, not to the cloud. That is a fundamentally different bet from both Neural and Line 6. Kemper is keeping the processing local and deterministic rather than depending on cloud infrastructure. Whether that matters to you depends on how you feel about needing an internet connection to make a capture, but it is a real philosophical difference. And Liquid Profiling carries forward into 2.0, which means profiles can still get proper amp control behaviour by mapping the original amp’s gain and tone stack onto the profile. That was Kemper essentially admitting two years ago that the frozen snapshot was not enough on its own, and 2.0 doubles down on that by making the snapshot itself more detailed while keeping the interactive controls intact. MK1 units can play 2.0 profiles at lower resolution, so the install base is not left behind entirely.
TONEX sits a little differently, but the same pattern is there. IK’s AI Machine Modeling pitch is straightforward: compare the DI and processed signals, build a Tone Model, and get you to a usable result quickly. That is a huge reason TONEX took off. It is fast, practical, and gets a lot of sound into the box in a hurry. It also proves that for a lot of players, convenience wins. Nobody is sitting there asking for a philosophy lecture when they just want their favourite amp sound before rehearsal starts in twenty minutes.
It is also worth noting that Fractal exists in a completely different lane here. They have never done captures at all. The Axe-FX and FM series are pure modelling, and the recent AM4 is their most affordable unit yet while staying entirely in that camp. That is relevant because it reinforces the point that the modelling-only approach still has a committed audience, and it makes Line 6’s decision to build both lanes even more interesting by contrast. Fractal is betting that if you make the modelling good enough, you do not need captures. Line 6 is betting that you should not have to choose.
So where do I think Proxy could be strongest when it lands publicly? Not in the weird science projects. Not in the internet flexes. Not in some guy insisting he captured a 1959 amp on a rainy Tuesday and now the rest of us are obliged to care. I think Proxy has the best chance of being genuinely useful when it is pointed at real-world, repeatable guitar problems. Load-box captures. Preamps. Drives. A mic’d-up amp setup you want to drag from studio to stage without reinventing it every time. That all lines up pretty neatly with how Klein framed phase one, and it is a much saner starting point than trying to claim you have already solved every edge case in guitar gear.
It also fits the way Stadium itself is built. You have up to 48 dynamic blocks across four stereo paths, which means there is plenty of room for hybrid presets where a Proxy clone handles the core sound and the rest of the rig is still classic Helix territory: delays, reverbs, routing tricks, sidechains, pitch effects, whatever weird little ecosystem you have built for yourself over the years. That hybrid use case is more interesting to me than the usual all-or-nothing capture talk, because most guitarists are not looking for a museum exhibit. They want a reliable starting point they can actually use in a preset.
Where I would stay cautious is exactly where I stay cautious with every capture system. Weird fuzzes. Hyper touch-sensitive amps. Compressors where the feel is half the sound. Edge-of-breakup tones where the guitar volume knob is doing just as much work as your picking hand. Those are the places where capture tech tends to get exposed, and Neural’s own Capture V2 messaging more or less says as much by specifically calling out touch-sensitive devices as an area of improvement and dedicating entirely new device categories to the problem.
That is why I do not really care whether Proxy is more AI than the competition or less AI than the competition. That whole conversation feels like the gear equivalent of arguing which energy drink contains the most lightning. I care whether it is effective. I care whether it gets me to the sound faster. I care whether it still feels convincing when I stop playing polished demo riffs and start playing like myself, which is always a more brutal test.
The smartest thing about the Stadium strategy is that Line 6 do not seem to be forcing one method to solve every problem. Agoura can be the lane for players who want deep, physically informed amp behaviour and a model that stays interactive when they start fiddling with controls. Proxy can be the lane for players who already know the exact sound they want and would rather clone it than rebuild it. Hype can split the difference when you want the shape of the real thing but not every wart. That is a much more sensible approach than pretending one single technology is going to dominate every use case forever.
And if Proxy is good in version one, that probably will not be the end of the story anyway. This part of the market is moving fast. Neural went from the original on-device capture process to cloud-trained V2. Kemper went from the original profiling engine to a 100,000-frequency-point system that offloads to your computer and ditches the refining phase. Everyone is still iterating. Klein even said work on phase two is already underway. So even if Proxy arrives as merely good at first, the interesting question is what it looks like after a few firmware cycles and a year of Line 6 feeding it more data and more edge cases.
My read is pretty simple. Helix Stadium does not look like it is trying to become a Quad Cortex clone, a TONEX clone, or a Kemper clone with different fonts. It looks like Line 6 are trying to build a platform where modelling and cloning support each other instead of fighting for the same patch of turf. If they pull that off, Proxy will not just be another checkbox feature arriving late to the party. It will be one half of a genuinely smarter system, and that is a lot more interesting than another round of capture chest-beating.