Slot chain
Up to 8 slots: each a NAM model (amp, pedal), an IR (cab, room, reverb, delay) or the built-in 2x12 cab. Turn on/off, reorder, per-slot level and mix. New slots go to the right place: pedals before the amp, cabs after.
A standalone Neural Amp Modeler (NAM) guitar amp for Android: plug a guitar into a USB audio interface on your phone and play through NAM amp and pedal models, IRs and a tone stack. No computer needed. Open source (MIT).
Android · NAM · USB audio interface · Open source
Your phone becomes a guitar amp you carry in your pocket.
Plug your guitar into a USB audio interface (e.g. Focusrite Scarlett 2i2) on your phone and your headphones into the interface: HandNAM runs NAM models, learned from real amps and pedals, on the phone's own CPU. No computer, network or cloud. The audio engine is written from scratch in C++; the UI is in English and Turkish.
How it works
Every step from guitar to ear runs on the phone, and is measured.
guitar → USB audio interface → gate → NAM / IR slot chain → bass / mid / treble → level → headphones
Up to 8 slots: each a NAM model (amp, pedal), an IR (cab, room, reverb, delay) or the built-in 2x12 cab. Turn on/off, reorder, per-slot level and mix. New slots go to the right place: pedals before the amp, cabs after.
Search the TONE3000 catalog in the app, see your favorites and downloads; tap a model to download it and add it to the chain. Each tone's creator and license are shown. Powered by TONE3000.
Full-duplex audio with Oboe, exclusive (MMAP) mode where available, audio thread pinned to big cores, two-stage partitioned convolution for long IRs. Keeps playing with the screen off.
NAM's new A2 (Slimmable) architecture is very light on phones: about 9 % of one big core per 96-sample block, against about 31 % for a standard WaveNet. The app puts A2 models first in search.
Test results
Samsung Galaxy A36 + Focusrite Scarlett 2i2 (USB-C), 48 kHz.
| Measurement | Value | Note |
|---|---|---|
| Guitar-to-ear latency | 10–14 ms | average 11.8 ms, exclusive (MMAP) mode |
| A2 model, one-core load | ~9 % | 96-sample block |
| Standard WaveNet, one-core load | ~31 % | large models may stutter |
| Worst audio block with long IRs | 80–91 % | 108–128 % before the tail moved to a background thread |
| Phone without MMAP (Tecno) | shared mode | higher latency, shown in the status line |
Screenshots
Main screen, Settings and open-source credits.



Open source
HandNAM is built on these projects; every license can be read in the app.
| Component | License |
|---|---|
| NeuralAmpModelerCore (Steven Atkinson) | MIT |
| Eigen | MPL-2.0 |
| nlohmann/json | MIT |
| Oboe (Google) | Apache-2.0 |
| HandNAM | MIT |
Status
HandNAM 0.5.0 is being prepared for an open-source release: the source code will be on GitHub soon under the MIT License. You need Android 8+ (64-bit) and a class-compliant USB audio interface; a free account for TONE3000. The app contains no TONE3000 key; TONE3000 models belong to their creators.