The language thatthinks like a brain.
Neuro puts tensors, autodiff, and GPU execution in the language itself rather than in a library layered on top of it. The core language compiles to native code today. The AI stack ships phase by phase.
// Neuro — AI-first, compiled
@grad
func model(x: Tensor<f32, [784]>) -> f32 {
val h = x.matmul(w1).add(b1).relu()
val y = h.matmul(w2).softmax()
loss(y, labels)
}
func main() -> i32 {
train(model, dataset)
return 0
}Neural architecture
Built for the AI era
The core language compiles and runs today. The AI-native layer (tensors, autodiff, GPU) arrives phase by phase. Every card below says which of the two it is.
Born for AI
Tensors, neural networks, and IR-level autodiff live in the core language, where the compiler can optimise them the way it optimises arithmetic. A library cannot reach that deep.
Lightning Fast
An LLVM backend compiles your code ahead of time into a native binary. Nothing interprets it at runtime, and no garbage collector stops it to clean up.
Joy to Write
Type inference, generics, traits, closures, enums, exhaustive pattern matching, and Option/Result in the prelude. All of it works today.
GPU Native
The same source targets NVIDIA and AMD through standard MLIR GPU dialects, so you do not hand-write CUDA once per vendor.
Shape-Safe
Static-rank tensor types catch shape mismatches at compile time, before they turn into a 3 AM debugging session. The MLIR tensor path carries this.
@grad
func f(x: f32) -> f32 {
relu(x)
}Zero-Cost Abstractions
Ownership semantics and fully monomorphized generics lower to the same machine instructions you would have written by hand. The abstraction disappears at compile time.
Quick setup
From source to native binary
Clone the repo, build the compiler, compile your first program. You need the Rust toolchain and nothing else.
Read the full guide- 1
Install Neuro
Build from source with Cargo
terminalbashgit clone https://github.com/PanzerPeter/Neuro.git cd Neuro cargo build --releaseRequires the Rust toolchain
- 2
Compile to a native binary
Type-check, then build with neurc
hello.nrneurofunc main() -> i32 { return 0 }cargo run -p neurc -- compile hello.nr
Write your first Neuro program
The compiler builds from source in a few minutes. The installation guide takes you from a git clone to a running binary.