A specialized local LLM orchestrator for Apple Silicon designed to execute Qwen2.5 models efficiently via the MLX framework.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install mlx-brain
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-brain 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 Brain?
MLX Brain is a high-performance orchestration skill designed specifically for developers using Apple Silicon hardware. It leverages the MLX framework to provide seamless local inference for state-of-the-art Large Language Models. By integrating this into your set of Openclaw Skills, you can offload compute-heavy AI tasks from the cloud to your local MacBook, ensuring data privacy and reducing latency.
This skill acts as a bridge between agent-based automation and local hardware, allowing users to invoke powerful models like Qwen2.5-7B for general instructions or coding tasks. It is built to work within distributed environments where a central agent can trigger local compute nodes for specialized processing.
MLX Brain Use Cases
- Offloading LLM inference from cloud agents to local MacBook Pro hardware to save costs.
- Executing sensitive coding tasks locally using the Qwen2.5-Coder model.
- Building low-latency AI workflows within the Openclaw Skills ecosystem.
- Automating local system tasks using structured JSON inputs and LLM-driven logic.
How MLX Brain Works
- The user or a remote agent triggers a command through the clawdbot node invocation system.
- The request is routed to the specific MacBook node where MLX Brain is installed.
- The skill activates the pre-configured Python virtual environment containing the MLX libraries.
- The run script loads the selected model (Qwen-Instruct or Qwen-Coder) into unified memory.
- Local inference is performed on the GPU/Neural Engine, and the response is piped back to the agent.
MLX Brain Setup
To use this within your Openclaw Skills, ensure you have a virtual environment at $HOME/mlx-env with MLX installed.
# Verify your local installation with a test prompt
$HOME/mlx-env/bin/python3 $WORKSPACE/misskim-skills/mlx-brain/run.py "Hello, how are you?"
# Invoke remotely via clawdbot
clawdbot nodes invoke --node "MacBook Pro" --command "system.run" \
--params '{"command":"$HOME/mlx-env/bin/python3 $WORKSPACE/misskim-skills/mlx-brain/run.py \"Hello?\""}'
MLX Brain Data Schema & Taxonomy
The skill accepts both raw string arguments and structured JSON data for more complex interactions:
| Parameter | Type | Description |
|---|---|---|
| prompt | string | The main text or instruction to process. |
| model | string | The identifier for the model (e.g., 'qwen' for instruct, 'coder' for coding). |
| --json | flag | Forces the script to parse input as JSON and return structured output. |
Models supported:
- qwen: Qwen2.5-7B-Instruct-4bit
- coder: Qwen2.5-Coder-7B-4bit
MLX Brain Advanced Features
- Support for remote node execution allowing your MacBook to act as an AI server for other Openclaw Skills.
- Optimized 4-bit quantization for high-speed inference on Apple Silicon unified memory.
- JSON-mode integration for programmatic piping between different automation nodes.
- Dynamic model switching between general instruction and specialized coding assistance.
SKILL.md
Loading
Related Openclaw Skills
Web Interface Guidelines Reviewer
A specialized tool for auditing web UI code against professional interface guidelines and accessibility standards.
Web Deploy GitHub Pages
Automate the entire lifecycle of creating, building, and deploying professional static websites to GitHub Pages.
UI/UX Pro Max
An advanced design intelligence skill for building, auditing, and implementing polished user interfaces and design systems.
Ralph Loop v2
A high-performance 3-tier AI agent methodology designed for autonomous software implementation with rigorous quality control and architectural integrity.
Playwright Testing (MiniPC)
An automated testing skill for verifying web applications and games using headless Playwright on remote MiniPC nodes.
PostgreSQL Performance Optimizer
A comprehensive optimization guide and toolkit for improving PostgreSQL query speed, schema efficiency, and connection management.
Featured*
