---
name: whisper-local-safe
description: Local audio transcription via whisper.cpp server on the LAN. Use when transcribing voice notes or audio files with the local whisper-server (no external API), with automatic endpoint fallback and plain text output.
---

# Whisper Local Safe

Use this skill to transcribe local audio files against:

- `http://192.168.153.113:18801`
- `http://192.168.153.117:18801`

## Run

```bash
skills/whisper-local-safe/scripts/transcribe-local.sh /path/to/audio.m4a
```

Optional env vars:

- `WHISPER_BASE_URL` (override endpoint)
- `WHISPER_MODEL` (default `base`)
- `WHISPER_LANGUAGE` (default `auto`)
- `WHISPER_OUTPUT` (path for transcript output)

## Behavior

- Tries `whisper.cpp` endpoint first: `/inference`
- Falls back to OpenAI-style endpoint: `/v1/audio/transcriptions`
- Returns transcript text to stdout and optional output file