Since you've already built whisper.cpp with Intel oneAPI/SYCL on your Gentoo system, the easiest way to run Whisper as a background service is to use the built-in whisper-server executable. It exposes an HTTP API that is compatible with many Whisper/OpenAI clients.
Start the server manually
If you compiled with SYCL support, it will automatically use your Intel Arc GPU when available.
Verify it's running:
or
Create a systemd service
Create:
Reload and enable it:
Check status:
View logs:
Test transcription
Or, for OpenAI-compatible clients, start the server with the appropriate request path, for example:
This allows clients that expect the OpenAI Whisper transcription endpoint to connect directly.
Given your Core Ultra 9 185H + Intel Arc GPU + oneAPI/SYCL setup, this is the recommended configuration for fast local speech-to-text without requiring Python or PyTorch.