Evolve neural networks to solve real-world problems.
Website · Documentation · Research
Quale is a neuroevolution platform that evolves connectome-based neural networks using NEAT (NeuroEvolution of Augmenting Topologies). Write a .quale file describing your domain - sensors, actuators, physics, scoring - and the engine evolves brains that learn to solve it.
No training data. No gradient descent. No labeled datasets. Just selection pressure and millions of generations.
| Repo | Description |
|---|---|
| Engine | Core evolution engine, bytecode VM, parser, CLI. Go + C (CUDA/OpenCL) |
| App | Desktop GUI for running experiments. Wails + Svelte |
| Frontend | quale.dev website, research papers, wiki docs. Astro + Starlight |
| VSCode | VS Code extension - syntax highlighting, snippets, comment system |
| Run | quale.run - browser-based experiment viewer and brain inspector |
- Domain-specific language -
.qualefiles express entire simulations: world topology, perception, physics, state machines, fitness scoring. Zero Go or C required. - Bytecode VM - One source file compiles to bytecode that runs identically on CPU and GPU. No code duplication.
- GPU acceleration - CUDA (NVIDIA) and OpenCL (Intel/AMD) backends evaluate thousands of brains in parallel.
- Connectome architecture - Brains have regions, plasticity, and gain modulation inspired by computational neuroscience.
- Human Factors - Safety modelling on a route topology with entity interactions, state machines, and dynamics
- Network Security - Intrusion detection with per-record fitness and classification scoring
- Survival - Grid-world agents with directional sensing, resource consumption, and foraging
