Built on Vulkan compute — every vendor, every platform:
INFRASTRUCTURE
GPU-native.
Quantum-resistant.
High Performance Computing
One compute engine runs on every GPU vendor — NVIDIA, AMD, Intel, Qualcomm. Cross-platform. Zero vendor lock-in. Ship a single binary.
Artificial Intelligence
End-to-end training and inference on the Oa compute engine. 77K tok/s. Compute graphs with 4.56x replay speedup. No Python, no PyTorch.
Vision
Hardware video decode. GPU image transforms. Every frame arrives as a normalised BF16 tensor — ready for inference without a copy.
Mission
Civilization began when humans traded instead of raided. We're ensuring it continues.
The ability to transact is the ability to exist. We're building infrastructure that ensures it remains open for the next century.
We are the realm.
We study problems that shouldn't be solvable. Then we solve them.
GPU-native training at 77K tokens per second. Byte-level language models running end-to-end in C++ on the Oa compute engine. 100% GPU utilization at 10% VRAM.
1.26 million Dilithium-3 verifications per second. Post-quantum settlement with 25ms finality.
“We don't use frameworks. We write them.”
— Internal memo, Q4 2024
Performance
that breaks benchmarks.
GPU-Native ML
77K tok/s training. Byte-level LLMs on the Oa compute engine. Zero framework overhead.
Post-Quantum
Dilithium-3 signatures. 1.26M verifications per second on GPU.
Unified gRPC API
One endpoint. Bidirectional streaming. Post-quantum TLS. TypeScript SDK.
Zero Dependencies
No Python. No PyTorch. No Node.js. Single binary deployment.
Realm-sdk
// GPU-native inferenceconst response = await realm.generate({model: "reallm-9.6m",prompt: "Once upon a time",temperature: 0.7,maxTokens: 256,});// → 77K tok/s, constant-time inference
Built for developers
C++ / Oa Engine core. TypeScript SDK. gRPC API. Full documentation with examples.
The Realm is Real
The sovereign layer.
Post-quantum cryptography. GPU-native execution. Built for the next century of global finance.
