NeuralAmpModelerPlugin: a VST3 and AudioUnit host for Neural Amp Modeler captures

What the plugin actually solves for a Neural Amp Modeler user

Neural Amp Modeler produces models of real amplifiers, and those models are only useful if something can load them in the place where you record. The plugin is that loader. It is a VST3 and AudioUnit plug-in for Neural Amp Modeler, built with iPlug2, and it exists so that a capture can sit on a track in a DAW the same way an amp sim does.

The intended user is a guitarist or recording engineer who already works inside a host. The README is explicit that the standalone build is the weaker path: the I/O for the standalone does not inherit the stability of most plugin hosts, so complex routing is better handled by the VST3 or AU version inside a host. If your workflow is a DAW session with a fixed input and output, the plugin is the right shape. If you wanted a self-contained practice app with flexible routing, the README itself steers you away from the standalone.

There is a second audience the README addresses indirectly: people who do not want to compile anything. The repository points to Gateway, a fork of this repo, available at neuralampmodeler.com/users, for pre-built installers. That is the path the README offers for anyone who wants binaries without running the build scripts.

How the iPlug2 wrapper and the NAM core fit together

The repository is split so that the audio engine and the plug-in shell are separate. NeuralAmpModelerCore sits at the top level next to the NeuralAmpModeler directory, and the README describes this repository as a cleaned up version of the original iPlug2-based NAM plugin, with refactoring to follow practices recommended by the iPlug2 developers. The practical consequence of that split is that model handling lives in the core, while the plug-in shell handles host integration, parameters and the GUI.

The shell is iPlug2, which is what produces VST3 and AudioUnit from one codebase. Other entries at the top level confirm the multi-target setup: common-mac.xcconfig, common-win.props, common-ios.xcconfig and common-web.mk are per-platform build configuration files, and TemplateProject and REAPER sit alongside them. The README states the plugin could also support AAX, CLAP, Linux and iOS soon, which is consistent with a build system that already carries iOS and web configuration files while only shipping Windows and macOS support today.

AudioDSPTools and eigen are present as dependencies, and iPlug2 is a submodule, which is why the repository carries a .gitmodules file and a setup_container.sh at the root. None of this is documented in the README beyond the build scripts pointer, so anyone reading the tree has to infer the dependency chain from the directory names.

Building NeuralAmpModelerPlugin from the scripts

The README does not give a command line. It says there are build scripts in NeuralAmpModeler/scripts/ and that the workflows under .github/workflows show how to do it. That is the documented route: read the script or the workflow, then run it. Because iPlug2 is a submodule, a plain clone leaves that directory empty, and the root-level setup_container.sh exists for the container path.

Start by cloning with submodules so iPlug2 and the other dependencies are present.

git clone --recurse-submodules https://github.com/sdatkinson/NeuralAmpModelerPlugin.git
cd NeuralAmpModelerPlugin

Next, look at what the scripts directory offers for your platform before running anything, since the README does not name individual scripts.

ls NeuralAmpModeler/scripts/
ls .github/workflows/

On Windows, the top-level common-win.props is the shared property sheet the project files reference. On macOS, common-mac.xcconfig plays the same role. The README does not state which script to run for which target, so the workflow files are the authoritative example of the sequence, and the build badge at the top of the README points at build-native.yml.

If you would rather not build, the README's own answer is Gateway at neuralampmodeler.com/users. That fork is described as a way to get a pre-built installer from this repo without compiling.

The startup crash on Windows hybrid graphics

The README has a Rough edges section, and the graphics entry is the one most likely to bite a new user. If NAM crashes before the GUI comes up, the documented cause is an unsupported graphics configuration, specifically a dedicated graphics card such as an nVIDIA GPU combined with integrated CPU graphics on a Windows system. The fix the README gives is to open the control panel, select NAM or your DAW, and force it to use the graphics card.

That is a real limitation, not a setup detail. It means the plug-in can fail at launch on a machine that meets the stated platform requirement of Windows 10 64-bit or later. The README also asks anyone who understands the problem better to open an Issue, which is an admission that the cause is not fully characterised. If you are deploying this across several Windows machines with mixed graphics, budget time for that failure mode.

The second documented rough edge is standalone I/O. The README says it does not inherit the stability of most plugin hosts and is sparser on features, and it recommends the VST3 or AU plugin inside a host for complex routing. Linux is a third boundary: the README states the plugin supports Windows and macOS, and points Linux users to a separate LV2 plugin by mikeoliphant rather than claiming Linux support here.

How this differs from a commercial amp sim plugin

The obvious alternative for a guitarist is a commercial amp simulator with a fixed library of amps and cabinets. The difference in approach is where the models come from. A commercial sim ships its own curated set of amp voicings and you choose among them. This plugin ships no amps at all; it is a host for Neural Amp Modeler captures, so the sound you get depends entirely on which models you load and where you obtained them.

That distinction has practical consequences. A commercial sim is a closed system: one installer, one licence, one preset browser, and support when something breaks. This plugin is a loader, so the model files are your responsibility, and so is their provenance. The README does not document a model browser or a bundled library, and the Models/ directory in the tree is not described in the README. If you want to sit down and play without sourcing captures first, a commercial sim removes a step that this plugin leaves to you.

The other real alternative is the standalone build of this same project, but the README treats that as the fallback rather than the alternative, precisely because its I/O is less stable than a host's.

Maintenance cadence, licence and upgrade cost

The repository is not archived and the last push was on 2026-08-23, so the project is being touched. Releases tell a less even story. v0.7.13 shipped on 2024-12-21, then v0.7.14 and v0.7.15 both landed on 2026-06-02, roughly eighteen months later and within hours of each other. That pattern suggests bursts of release activity rather than a steady cadence, which matters if you are pinning a version for a studio machine: expect long quiet stretches between tags.

The licence is MIT, which is permissive and places few obligations on how you redistribute the plugin. Two caveats are worth stating without giving legal advice. First, MIT covers the plugin source; it says nothing about the licence of the Neural Amp Modeler models you load, which come from elsewhere and may carry their own terms. Second, the README points at Gateway, a fork, for pre-built installers, so a binary you download may not correspond to a tag in this repository.

Upgrade cost is mostly build cost. Because iPlug2 is a submodule and the tree carries eigen and AudioDSPTools, moving between versions means updating submodules and rebuilding, not swapping a file. The README does not document rollback or a migration path between versions, so keep the installer or build output for the version you are currently running.

Editorial conclusion

Adopt this plugin if you already use Neural Amp Modeler captures and want them inside a VST3 or AudioUnit host on Windows 10 or macOS 10.15 and later, and if you are willing to build from the scripts in NeuralAmpModeler/scripts/ or take a pre-built installer from the Gateway fork. Do not adopt it if you are on Linux, where the README points at mikeoliphant's LV2 plugin instead, or if you need AAX or CLAP today, which the README lists only as possible future targets. Before committing, verify that your graphics configuration does not trigger the startup crash described under Rough edges, and confirm the licence terms of any model files you load, since the MIT licence covers the plugin code and not the captures.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. Releases
5. sdatkinson/NeuralAmpModelerPlugin on GitHub