name: openai-whisper
description: Local speech-to-text with the Whisper CLI (no API key).
homepage: https://openai.com/research/whisper
metadata:
{
"openclaw":
{
"emoji": "🎤",
"requires": { "bins": ["whisper"] },
"install":
[
{
"id": "brew",
"kind": "brew",
"formula": "openai-whisper",
"bins": ["whisper"],
"label": "Install OpenAI Whisper (brew)",
},
],
},
}
Agent for performing local speech-to-text transcription and translation using the OpenAI Whisper CLI.
- Tooling: `whisper` CLI.
- Execution Mode: Terminal execution.
- Capability Scope: Local, offline audio transcription and translation into various formats (txt, srt, etc.).
- MUST provide exactly one valid input audio file path.
- MUST specify the target `--model` (e.g., `tiny`, `base`, `small`, `medium`, `large`, `turbo`). If omitted, defaults to `turbo`.
- MUST expect and allow the model to download to `~/.cache/whisper` on the first run.
- MUST use `--task translate` if the goal is to transcribe non-English audio into English text.
- MUST specify `--output_format` (e.g., `txt`, `srt`, `vtt`) and `--output_dir`.
- AVOID running large models if system resources are constrained or if speed is prioritized.
1. Identify the absolute path to the audio file.
2. Determine the appropriate model size, task (transcribe or translate), and desired output format.
3. Construct the execution command:
- `whisper --model --output_format --output_dir [--task translate]`
4. Execute the command and wait for completion.
- Output the exact path to the generated transcript file.
- Do not attempt to summarize massive transcripts in the output unless specifically requested; point the user to the file.
By x10aix schedule Updated 5/21/2026