Start with the environment doctor. It checks the selected profile without installing packages or changing the environment.
screencastgen doctor --profile auto
Automatic setup selects local-gpu only on Linux or WSL2 when nvidia-smi can
access an NVIDIA GPU. Native Windows, macOS, and non-GPU Linux default to the
remote-client path. Run python3 scripts/setup.py --check to inspect the
decision before installation.
Pass the server URL explicitly so the doctor can validate /health:
screencastgen doctor \
--profile remote-client \
--server-url http://gpu-vm:8100
This commonly means the expected cuDNN runtime is unavailable. Follow WhisperX CUDA troubleshooting to inspect the active Torch, CUDA, and cuDNN libraries.
Pipelines intentionally reuse their status file and completed chunks. Pass
--clean to discard resumable state, or choose a different --output-dir and
--status-file for an independent run.