Accelerating the future.
Today.

Platform

Welcome to the Library.

Audio

Signal processing and audio analysis on GPU. Speech recognition, synthesis, and audio machine learning in a unified framework.

Core

The foundation. Memory management, threading, matrix operations, and file I/O in a single library. Built for performance across all hardware.

ML

Machine learning infrastructure. Layers, optimizers, and autograd for training and inference. Deterministic results across vendors.

Vision

Image and video processing. Decode, transform, and prepare visual data for analysis. From camera to model in one pipeline.

The Library. Every Vendor. Vulkan Native.

Cross-vendor compute infrastructure measured on real hardware. 640K samples/s on RTX 5090, 213K on Intel iGPU.

Fashion-MNIST at 86%+, text generation at 100% accuracy. One binary, every accelerator.

Documentation HPC AI

Mission

Progress began when humans harnessed energy instead of endured scarcity. We're ensuring it continues.

The ability to compute is the ability to create. We're building infrastructure that runs everywhere, on everything, for the next century.

We are the realm.

Research Laboratory

We study problems that shouldn't be solvable. Then we solve them.

Machine Learning

Controlled C++ and Python workloads verify parameter parity, convergence, generation, checkpoint reload, and host-versus-GPU timing on the same kernels.

Core Compute

Kernel routes, synchronization, memory ownership, and device capabilities are measured directly across integrated and discrete Vulkan systems.

Architecture claims are published with the hardware, software, workload, and measurement protocol required to reproduce them.

For Developers

Let the numbers speak for themselves.

Metric	Wall / step	GPU / step	Throughput	Evaluation
Transformer · Byte	9.24 ms	7.52 ms	111K token/s	92.8% acc
Transformer · BPE	9.59 ms	7.88 ms	296K byte/s	0.0107 bits/byte
MoE Transformer · Byte	13.55 ms	10.11 ms	75.6K token/s	92.8% acc
Mamba-3 · Byte	35.56 ms	33.57 ms	28.8K token/s	93.0% acc

Oa-library

OaFnMatrix::SetRngSeed(20260714);
auto a = OaFnMatrix::RandN({1024, 384});
auto b = OaFnMatrix::RandN({1536, 384});

// Records C = A × Bᵀ into the current compute context.
auto c = OaFnMatrix::MatMulNt(a, b);

OaContext::GetDefault().Execute();
OaContext::GetDefault().Sync();

The Library · GPU-First · Vulkan.

C++ core library, Python bindings, and capability-driven execution across integrated, discrete, and mobile Vulkan devices.

Documentation

The Library · GPU-First · Vulkan

Built with passion.

Compute, machine learning, vision, audio, and cryptography through one capability-driven runtime for integrated, discrete, and mobile GPUs.

Read the Docs

Accelerating the future.Today.