v3.1.0 Released! 🚀

The AI-Native
Programming Language.

Ferrite is a compiler-first, expression-oriented programming language built specifically for Artificial Intelligence and Machine Learning. Native tensors, compile-time shape verification, scope-based native memory, and high-performance binaries.

Try in Browser Explore the Docs

tensor_multiply.fe


import "math";

// Compile-time validated tensor shapes
param features: Tensor<float, (100, 784)> = ones();
param weights: Tensor<float, (784, 10)> = rand();

infer {
    // Built-in matrix multiplication operator
    keep logits = features @ weights;
    println("Output shape: " + str(shape(logits)));
}

Why Ferrite over Python?

Python is dynamically typed and relies on external runtime frameworks. Ferrite is built different.

The Old Way (Python)

❌ Runtime tensor errors
❌ Dynamic typing
❌ Framework-dependent

The Ferrite Way

✅ Compile-time tensor safety
✅ Strict Static typing
✅ Expression-oriented syntax
✅ Compiler-generated autodiff

Native Benchmarks (v3.1.0)

Thanks to the native LLVM backend introduced in v3.0.0 and perfected in v3.1.0, Ferrite consistently matches or beats Rust in pure math loop-unrolling and heavily outperforms Python across the board!

Benchmark	Ferrite	Python	Node.js	Rust	Go	Lua
ackermann	989 ms	2347 ms	841 ms	1059 ms	1095 ms	TBD
collatz	1009 ms	3771 ms	705 ms	745 ms	945 ms	TBD
fibonacci	862 ms	982 ms	714 ms	891 ms	892 ms	TBD
loop_sum	653 ms	887 ms	716 ms	937 ms	872 ms	TBD
mandelbrot	739 ms	1275 ms	1154 ms	773 ms	941 ms	TBD
string_concat	851 ms	884 ms	733 ms	772 ms	1003 ms	TBD

Built for Performance

🦀

Rust Foundation

No GC overhead. Built on top of safe systems memory.

📐

Shape Checking

Math errors are caught at compile time, not in production.

⚡

AOT Binaries

Ship standalone executables to run anywhere instantly.

The AI-Native Programming Language.