Project
NAM AUv3: Neural Amp Modeler
Neural amp models and cabinet IRs — real amp tone in any AUv3 host
What it is
NAM AUv3 is an iOS and Mac AUv3 plugin that loads Neural Amp Modeler (.nam) profiles and cabinet impulse responses (.wav) for real-time guitar amp simulation. Built on Swift 6 with a C++ DSP kernel and the vendored NAM Core (using Eigen) for neural inference on-device, it supports ConvNet, WaveNet, and LSTM model architectures and calibrates its input to match the level the model was trained at.
Who it’s for
Guitarists and producers using iOS and Mac DAWs like AUM, GarageBand, Cubasis, Logic Pro, or Loopy Pro who want real tube amp tone without external hardware.
Features
Neural amp modeling
Load any community-trained .nam model and play through a neural capture of a real amp, in real time.
Cabinet IR loading
Drop in any .wav impulse response to shape the cabinet, room, and microphone character.
3-band EQ
Shape your tone with a built-in low, mid, and high EQ stage after the model.
Noise gate
Tame idle hum and pickup noise with a configurable threshold gate before the model.
Normalized / raw / calibrated output
Match level to your host or preserve the true modeled output with three output modes.
Works in any AUv3 host
Runs as an Audio Unit extension on iOS and Mac — Logic Pro, AUM, GarageBand, Cubasis, Loopy Pro, and other compatible hosts.
Model and IR library
Browse, select, and delete models and IRs from the plugin's selector. Full-width rows on iPhone keep the whole file name readable.
Slimmable models
Adaptive model sizing trades a little quality for CPU headroom when a session needs it.
Fully accessible
VoiceOver labels every control — adjust knobs, toggle effects, switch output modes, and hear input and output levels announced while you play.
Tech
- Swift 6
- C++ DSP kernel
- Vendored NAM Core (Eigen, ConvNet, WaveNet, LSTM)
- Real-time audio
- AUv3 extension architecture
- iOS 16+ / macOS 13+
Links
A plugin, not another amp modeler
NAM AUv3 focuses on being a great host plugin rather than reimplementing the modeling. The neural inference is handled by the same vendored NAM Core used by the desktop Neural Amp Modeler project, so any model that loads in the desktop tool loads here too — thousands of free community captures are available at TONE3000 and ToneHunt. The plugin plays existing captures; it does not train new ones.
Workflow
- Open NAM AUv3 inside your host on iOS and Mac (AUM, GarageBand, Cubasis, Logic Pro, Loopy Pro).
- Pick a
.nammodel file from the built-in library — a neural capture of a real amp. - Optionally load a cabinet impulse response (
.wav) to shape the cabinet. - Dial in the noise gate, EQ, output mode, and input calibration.
- Play. The DSP runs in the audio thread with the host's real-time priority.
Why neural amp modeling
Neural amp modeling captures a real amplifier's response across its full operating range — clean, edge of breakup, and high-gain — rather than approximating it with a handful of EQ curves. A well-captured model reproduces how the amp reacts to your guitar's volume knob, your pick attack, and the pedals in front of it, in a way traditional amp sims struggle to match.
The model ecosystem
Because NAM AUv3 reads the standard .nam format, you get access to a growing library of free community models:
- TONE3000 — curated, downloadable amp captures and IRs.
- ToneHunt — a large community archive of models and impulse responses, free to download and use.
The same files you'd use on desktop work on iOS and Mac — no conversion, no retraining.
Built on NAM Core
The plugin ships the upstream NAM Core C++ inference engine (built on Eigen, with ConvNet, WaveNet, and LSTM architectures) inside the AUv3 extension. Models are resampled when the host's rate differs from the model's, and slimmable models can trade a little quality for CPU headroom. That keeps model loading and DSP behavior consistent with the desktop Neural Amp Modeler project while running entirely on-device.
Accessible by design
Every control has a VoiceOver label. Knobs are adjustable and announce their exact values, toggles announce their state, the model and IR rows speak which file is loaded, and the input and output meters announce their level while you play. The full library — browse, select, and delete — works without seeing the screen.