A local, 24/7 OpenAI-compatible API server for high-performance STT and TTS powered by the MLX framework on macOS.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install mlx-audio-server
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install mlx-audio-server 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 Audio Server?
MLX Audio Server is a specialized tool designed to bring lightning-fast, hardware-accelerated audio processing to Apple Silicon Macs. By leveraging the MLX framework, it provides an OpenAI-compatible API for speech-to-text and text-to-speech tasks, operating entirely on-device to ensure privacy and low latency. As part of the Openclaw Skills ecosystem, this server allows developers and AI agents to process audio files or generate speech without relying on external cloud providers.
The server runs as a persistent LaunchAgent, meaning it is always ready to handle requests from your local coding environment or automation scripts. It integrates seamlessly with popular MLX-optimized models, ensuring that your Mac hardware is utilized to its full potential for AI-driven audio workflows.
MLX Audio Server Use Cases
- Creating private, local transcriptions for audio and video recordings.
- Generating high-quality voiceovers for applications using local speech synthesis.
- Providing a local backend for AI agents within Openclaw Skills that require voice interaction.
- Reducing costs by replacing paid OpenAI Whisper or TTS API calls with local execution.
- Automating bulk audio processing tasks via command-line scripts.
How MLX Audio Server Works
- The server is installed and registered as a macOS LaunchAgent to ensure 24/7 availability.
- It loads optimized MLX models such as GLM-ASR for transcription and Qwen3-TTS for speech generation.
- Upon receiving an STT request, the system uses ffmpeg to normalize input audio into the required WAV format.
- The MLX framework processes the audio data using the Mac GPU or Neural Engine for high-speed inference.
- For TTS requests, the server converts text into speech and returns the path to the generated audio file.
- Standard OpenAI-compatible API endpoints allow for easy integration with existing AI tools and Openclaw Skills.
MLX Audio Server Setup
To set up the server on an Apple Silicon Mac, ensure you have Homebrew installed and run the following command from the skill directory:
bash ${baseDir}/install.sh
This process will install ffmpeg, jq, and the mlx-audio-server formula, then automatically start the background service.
MLX Audio Server Data Schema & Taxonomy
The skill manages audio conversion and model metadata as follows:
| Feature | Default Model / Format | Description |
|---|---|---|
| STT Model | mlx-community/glm-asr-nano-2512-8bit | Optimized for fast and accurate local transcription. |
| TTS Model | mlx-community/Qwen3-TTS-12Hz-1.7B-VoiceDesign-bf16 | High-fidelity voice synthesis model. |
| Input Support | Audio/Video (via ffmpeg) | Automatically converted to WAV before processing. |
| Output Path | speech.wav | Default filename for generated speech in the specified output directory. |
MLX Audio Server Advanced Features
- Background service persistence via macOS LaunchAgent integration.
- Support for custom output directories for generated speech files.
- Full compatibility with OpenAI API specifications for STT and TTS.
- Native Apple Silicon optimization using the MLX framework for maximum throughput.
- Easy integration with other Openclaw Skills for automated media processing pipelines.
SKILL.md
Loading
Related Openclaw Skills
HackerNews Extract
A utility that extracts a HackerNews post, including the linked article and all threaded comments, into a single, well-formatted Markdown document.
Media Downloader
A smart media retrieval skill that downloads video and audio from various platforms directly into organized local folders.
Zhihu Hot List Monitor
A specialized monitoring tool designed to fetch, filter, and analyze the most popular trending topics and questions on Zhihu.
Xueqiu Hot Discussions Monitor
A monitoring tool for fetching real-time trending stock discussions and investment insights from the Xueqiu platform.
mlx-stt
A high-performance, local speech-to-text solution optimized for Apple Silicon using MLX and open-source models.
MLX TTS
A high-performance local text-to-speech engine for macOS that leverages Apple Silicon and MLX for fast, private audio synthesis.
METADATA
Featured*
