Source Available
Cross-platform AI inference runtime on Vulkan
One codebase across AMD, Qualcomm, Intel, and NVIDIA GPUs.
Context
Inference is moving into Physical AI
The shift
AI inference is expanding from server-hosted single-shot services into Physical AI's continuous control loops — sensor, inference, control. Real-time processing becomes a core requirement rather than an optimization.
The constraint
That requirement is met only by GPU-based pipelines that keep every stage, from sensor input through visualization, resident in GPU memory. Copies between stages are the bottleneck.
The gap
The cross-platform counterpart of this category is still emerging. Existing GPU pipeline stacks are largely single-vendor. vAiStrata is building that counterpart.
Components
An engine, and the pipeline around it
vAi Inference Engine
Vulkan compute inference engine. The same model runs on AMD, Qualcomm, Intel, and NVIDIA GPUs using each platform's stock driver, on both Linux and Windows.
vStream GPU Pipeline
Zero-copy pipeline for multi-sensor input. Decode, preprocessing, fusion, post-processing, and rendering execute in place without round-trips through host memory.
Validation
One codebase across desktop and edge
The same build runs on a discrete desktop GPU and on an ARM edge SoC with heterogeneous GPU/NPU execution.