01◈
Native desktop studio
Runs natively on Windows, macOS, and Linux. No browser tabs, no accounts, no subscriptions — just an app that lives on your machine.
⇢what you get
Everything you need, nothing you don't.
02✦
Guided training
Create a project, pick a model, import a dataset, and start training through a visual interface a novice can master.
⇢03◇
Reusable builds
Save complete training configurations as Builds. Create workflows, share them, reproduce experiments, and switch setups instantly.
⇢04⌬
Auto environment setup
Python detection, dependency installation, GPU detection, hardware compatibility — all the boring stuff, handled for you.
⇢05❖
Project management
Models, datasets, projects, training runs, builds, and settings — organized in one place, not scattered across folders.
⇢06▣
Radically local-first
Your datasets, checkpoints, and projects remain yours. Know exactly what went into your model — and what it costs the planet.
⇢the app
Coded by AI, designed by humans.
The methodology is honest about it: all the abstract planning, thinking, and mapping — every component, workflow, page, and aesthetic — is done by a human, extensively designed before a line of code exists. The implementation itself is written by AI, then verified by the developer before each release.
Here are examples of our (100% human-made) art.
Launch
File explorer
Export
Inspect
Loading
Error
the flow
From idea to model, step by step.
1
Create a project
Name it, choose a model, and import a dataset from the visual interface.
2
Configure a build
Set your hyperparameters once, save them as a reusable build.
3
Train & monitor
MAMA sets up the environment, finds your GPU, and runs the training.
4
Run your model
Use your finished model locally, fully aware of what went into it.
why
The philosophy.
"AI should be difficult because of the problems you're solving — not because of the tools you're forced to use."
Agency
MAMA exists to give people agency over AI. If you have reasonable hardware, you can create and run any kind of model with no prior AI experience — training on data you curated yourself, with no corporate data-stealing and no hidden costs.
Accountability
You know exactly how much energy, resources, and water go into running and training your models. Every component, workflow, page, and aesthetic was designed by a human — and every release is verified by the developer before it ships.
OPEN SOURCE, BY DESIGN
Built in the open. Verified by hand.
MAMA is free and Beerware-licensed (revision 42). The entire source is public on GitHub, so anyone can read it, audit it, and run it — no telemetry, no accounts, no lock-in. This project leans heavily on AI-generated code, and its developer owns that fact openly, rather than hiding it.
Python
PyWebView
PyTorch
TensorFlow
Windows
macOS
Linux
STANDING ON THE SHOULDERS OF OTHERS
- pytorch-template by @victoresque
- Tensorflow-Project-Template by @mgsalem
MAMA is released under the Beerware license (revision 42).
join in
Make it yours.
Fixing bugs, improving docs, designing UI, adding training backends — all welcome. New here? The Issues page has beginner-friendly tasks waiting for you.