LibreRT propose a SYCL-based programming model with extensions for real-time, e.g., with priority-based scheduling for CPU and GPU with preemption support, as well as approximate computing, which supports state-of-the-art mixed precision and GPU kernel perforation techniques. The LibreRT programming model builds on modern C++ semantics and advanced runtime systems, and is capable to target embedded GPUs. For the Operating System, LibreRT will target micro-edge devices with Miosix, an OS with higher responsivity and advanced scheduling support; deep- and meta-edge devices will rely on more traditional Linux distribution.

Finally, an autotuning layer will introduce adaptive, self-tuned methods that tune the approximation accuracy to meet the real-time constraints. In terms of Key Performance Indicators, the project aims at delivering schedulers with preemption latency.