Hi all,
Here’s something fun for the weekend.
For those who like using AI, here’s the mod-plugin-cookbook:
it lets MOD users easily build and push a working plugin into their
MOD units, starting from a plain-language description of what they
want to hear.
The procedure is simple:
- Discuss your sound design idea with the AI of your choice (Claude,
ChatGPT, Gemini, others — any AI that can fetch URLs works). Give
it the link to the cookbook. It will produce a single.mkfile
for you. - Go to our Buildroot Cloud Builder
and upload the.mkfile from the previous step. You’ll need a MOD
unit connected over USB to proceed. - Tell the Builder to Build, then Install the plugin.
That’s it. The plugin should now show up in the Web UI 
Bonus: when uploading, you can tick the “shareable persistent
link” option. The builder gives you back a URL that anyone else with
a MOD unit can click to install the same plugin on their device.
Nice way to share what you made here on the forum or pass to friends
— a step toward community-driven plugins before they make it to the
Beta store.
I’ve tested this with a Tube Screamer, a CE-2 chorus, and a
bidirectional filter so far — all built and ran first time on my
Dwarf, across Claude and ChatGPT.
This is just a proto for a bigger idea — if it gets traction here, we’ll invest more
resources into it.
The current single-.mk-file model is appealing because
it’s simple and self-contained (the dev never has to push code
anywhere, the file carries everything), but it does come with limits:
- One source file per plugin. Multi-file DSP code doesn’t fit
cleanly. - No external libraries beyond DPF. So no
fftw, no
libsamplerate, no other deps. - No custom pedal face (no modgui). Plugins use the “tuna can”
auto-generated UI on the unit. - No iteration. Editing means regenerating the whole
.mkand
re-uploading. Fine for trying ideas, painful for serious
development. - Practical size limit around ~1000 lines of DSP code before the
file gets unwieldy.
A “turn this into a real project” mode is the natural next step if
people want it — proper source tree, version control, ability to
iterate. I’d rather see what people actually do with the prototype
first before designing it.
In terms of FX creation, It doesn’t matter if you want to ask for a model of a specific pedal — Tube Screamer, CE-2, ProCo Rat — or have a more abstract
conversation about sound design (“a wavefolder that gets harsher as
it’s pushed”, “a chorus that doesn’t add lows”). That’s fully up to you and the AI you choose to work with.
Let me know how you feel about it. I think it’s very exciting
Best wishes
Gianfranco, aka the MOD Father
What I did is using the chat in VS Code (on Mac) with the Copilot AI agent available through a Github account. I’m thinking about subscribing to Claude though.