MIND Language

What is MIND?

MIND (Machine INtelligence Development) is a systems programming language designed specifically for AI and high-performance computing. NIKOLA is written entirely in MIND, leveraging its native tensor operations and zero-overhead abstractions.

Why MIND Instead of Python/Rust?

Key Features

Native Tensor Types

First-class tensor support with compile-time dimension checking:

// HalfKA NNUE architecture: 45056 → 1024 → 8 → 32 → 1
let ft_weights: tensor<i16, (45056, 1024)> = load_weights("ft.bin");
let features: tensor<i8, 45056> = extract_halfka(&position);
let accumulator = ft_weights.forward(&features);  // GPU-accelerated

GPU Kernels

Write GPU code inline with automatic backend selection:

fn gpu_batch_evaluate(positions: &[Board], batch_size: usize) -> Vec<i32> {
    let mut results = vec![0i32; positions.len()];
    on(gpu0) {
        parallel for i in 0..positions.len() {
            let features = extract_halfka(&positions[i]);
            results[i] = network_forward(&features);
        }
    }
    results
}

Zero-Cost Abstractions

High-level code compiles to optimal machine code:

// This high-level code...
let best_move = moves
    .filter(|m| is_legal(m))
    .max_by(|m| evaluate(m))

// ...compiles to hand-optimized assembly

Cross-Platform Compilation

Single codebase compiles to CUDA, Metal, ROCm, WebGPU, and CPU backends without modification. The MIND compiler automatically generates optimized kernels for each target.

MIND in NIKOLA

NIKOLA's 44 source files demonstrate MIND's strengths:

• GPU-Batched NNUE - Lazy SMP threads submit positions to GPU for batch evaluation (500M+ pos/sec)
• SPTT Hybrid Search - Dynamic switching between alpha-beta and GPU MCTS based on position
• GPU MCTS - Monte Carlo Tree Search with PUCT selection and virtual loss for parallelism
• Fortress Detection - CNN-based detection of defensive structures
• Move Generation - Magic bitboards with SIMD-optimized operations
• Distributed Hash - Lock-free transposition tables across cluster nodes

Getting Started with MIND

Install the MIND toolchain:

curl -fsSL https://mindlang.dev/install.sh | bash

irm https://mindlang.dev/install.ps1 | iex