use-local-whisper

Local voice transcription on Apple Silicon replaces the OpenAI Whisper API by executing whisper.cpp directly on-device. This keeps audio processing offline, free of network dependencies, and completely private by swapping the cloud path for the whisper-cli binary.

Running local whisper.cpp transcription requires whisper-cli, ffmpeg, and GGML model binaries installed and reachable. You must also configure the restricted launchd PATH so the background service can successfully locate and execute these binaries during voice transcription.

Local voice transcription works with NauggieClaw WhatsApp channels on macOS systems that already use the voice-transcription skill. Applying this skill swaps the existing cloud transcription path to use whisper-cli for on-device processing.

Verify local whisper.cpp transcription by building the project, restarting the launchd agent, and tailing logs. Confirming that transcripts appear as expected in WhatsApp group chats proves the whisper-cli execution path is functioning correctly.

Local whisper transcription operates entirely offline without any network connection. By replacing the OpenAI Whisper API with whisper.cpp and GGML models, audio processing remains local and independent of external network dependencies.

Local voice transcription fails when the launchd agent starts if whisper-cli, ffmpeg, or GGML model binaries are not reachable inside the restricted launchd PATH. Verifying dependency placement and PATH settings resolves these startup friction issues.