Hapi discovers hardware acceleration capabilities across 4 vendors: Apple Silicon (Metal, ANE, MPS), NVIDIA (CUDA, cuDNN), AMD (ROCm), and Intel (oneAPI). He reports VRAM, CPU core counts, unified memory, and available compute backends so Pantheon and Sirsi can route workloads to the fastest available hardware.
Detects Metal, ANE (Neural Engine), MPS (Metal Performance Shaders), unified memory
CUDA toolkit version, cuDNN, VRAM, GPU model detection
ROCm ecosystem, MI series detection
oneAPI, Arc GPU detection
Reports available backends so workloads route to fastest hardware
$ pantheon hapi
# Detect all available accelerators
$ pantheon hapi --json
# Machine-readable accelerator report
| Interface | Status | Detail |
|---|---|---|
| CLI | โ Full | pantheon hapi |
| Internal API | โ Full | Used by menu bar stats |
| Menu Bar | โ Active | Accelerator in stats panel |
| MCP | โ None | Planned |
This page is the canonical source for Hapi. Bundled offline copies auto-check for updates here.