Features and Benefits
Flexible Kernel Assortment
Kleidi libraries offer flexible kernels that enhance AI on frameworks, improving capability and accuracy for AI, accelerating performance, and reducing memory overhead.
Low Overhead for Developers
The KleidiAI library directly integrates into key AI frameworks—MediaPipe (via XNNPACK), llama.cpp, PyTorch (via ATen), ONNX Runtime and MNN—giving developers automatic performance benefits with no added overhead.
Unleashing Mass-Market AI Performance
As Kleidi libraries help optimize AI at the framework level, benefitting hundreds of workloads on billions of Arm-based devices. Developers can run models on Kleidi-optimized frameworks to achieve top performance.
Enabling Generative and Agentic AI at Scale
Kleidi simplifies and accelerates deployment of demanding AI inference workloads on Arm. The KleidiAI library delivers best-in-class performance for GenAI and agentic AI deployed from cloud datacenters to constrained devices at the edge.
Optimizing AI for Everyone—Everywhere
Kleidi enables easy optimization across Arm Neoverse, Arm Cortex-X, and Cortex-A CPUs. Its performance libraries leverage technologies like Arm Neon, Arm Scalable Vector Extensions (SVE), and Arm Scalable Matrix Extensions (SME), to accelerate AI functions on Arm architecture.
Concise, Efficient, and Light
The libraries are lightweight and concise, with no dependencies or binary release, and avoid duplicating memory allocation or multithreading implementation—making them easy to integrate into framework codebases efficiently.
