MLX Whisper for Openclaw

A local, high-performance speech-to-text engine optimized for Apple Silicon that eliminates the need for cloud APIs.

kevin37li
v1.0.0
Jan 25, 2026
1
4.7k
0

Install & Download

1. ClawHub CLI

The fastest way to install a skill directly from the registry.

npx clawhub@latest install mlx-whisper

2. Manual Installation

Copy the skill folder to one of these locations

Global
~/.openclaw/skills/
Workspace
<project>/skills/

Priority: Workspace > Local > Bundled

3. Prompt Installation

Copy this prompt to OpenClaw to install it automatically.

Help me install mlx-whisper using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).

Prefer to download?

Get the raw skill files in a ZIP archive.

What is MLX Whisper?

MLX Whisper is a specialized skill designed to leverage the power of Apple's MLX framework for lightning-fast audio transcription and translation. By running models locally on M-series chips (M1, M2, M3, M4), it provides a privacy-first approach to processing voice data without recurring costs or internet dependencies. This integration within the Openclaw Skills ecosystem allows developers to automate complex transcription tasks directly on their hardware.

The skill supports a wide range of model sizes, from the lightweight tiny model for rapid processing to the large-v3-turbo model for near-perfect accuracy. Whether you are generating subtitles for video or transcribing research interviews, this skill offers a professional-grade solution for macOS-based AI automation.

MLX Whisper Use Cases

  • Transcribing voice notes or interviews into clean text files for documentation.
  • Generating SRT or VTT subtitle files for video production workflows.
  • Translating non-English audio content into English text locally.
  • Automating bulk audio processing pipelines using Openclaw Skills on Apple Silicon Macs.

How MLX Whisper Works

  1. The user provides an audio or video file as input to the skill.
  2. The skill invokes the MLX-optimized Whisper model, which utilizes the Mac's GPU and Neural Engine.
  3. On the first run, the necessary model weights are securely downloaded from HuggingFace and cached locally.
  4. The audio is processed frame-by-frame to generate timestamped text or translations.
  5. The final result is exported in the requested format (TXT, SRT, etc.) to the local file system.

MLX Whisper Setup

To use this skill, ensure you have an Apple Silicon Mac and the necessary package installed via pip:

pip install mlx-whisper

You can then run a transcription using the following command structure:

mlx_whisper /path/to/audio.mp3 --model mlx-community/whisper-large-v3-turbo

MLX Whisper Data Schema & Taxonomy

The MLX Whisper skill manages data through local caching and structured file exports. It does not rely on external databases, ensuring all processed information stays on your device.

Data Type Description
Model Cache Weights are stored in ~/.cache/huggingface/ for persistent offline access.
Output Formats Supports standard text (.txt) and subtitle (.srt) file extensions.
Metadata The skill tracks language hints and task types (transcribe vs translate) during execution.
Openclaw Skills Integration Results can be piped to other local tools or saved to specific project directories.

MLX Whisper Advanced Features

  • Support for the large-v3-turbo model, offering the best balance of speed and quality for Openclaw Skills users.
  • Specialized translation task mode to convert foreign speech directly to English.
  • Language identification and hinting to improve accuracy in multi-lingual environments.
  • Zero-latency local processing that bypasses the need for any external API keys or cloud subscriptions.

SKILL.md


Loading

Related Openclaw Skills

METADATA

Github Stars: 0
forks: 0

Featured*