← Back to Skills Marketplace
Mlx Apple Silicon Mlx
by
Twin Geeks
· GitHub ↗
· v1.0.1
· MIT-0
124
Downloads
0
Stars
2
Active Installs
2
Versions
Install in OpenClaw
/install mlx-apple-silicon-mlx
Description
MLX-powered local AI — run LLMs, Stable Diffusion, speech-to-text, and embeddings natively on Apple Silicon via MLX. Ollama uses MLX for LLM inference, mflux...
Usage Guidance
This skill appears to be what it claims (a guide for running MLX-based local services), but take these precautions before proceeding: (1) Inspect the PyPI package (ollama-herd) source on GitHub before running pip install — PyPI packages run arbitrary code on install. (2) Verify what the 'uv' tool is, where it comes from, and whether it will download additional binaries/models. (3) Be aware that installing/starting 'herd' and 'herd-node' will run local network services (localhost:11435) and create files under your home (metadata lists ~/.fleet-manager/*); review those config and log paths for sensitive content. (4) Use a Python virtual environment or isolated machine if you want to limit blast radius. (5) Confirm the 'no automatic downloads' claim: model downloads are often large and may need explicit confirmation — check the herd/ollama-herd docs for exact behavior. If you want a lower-risk evaluation, provide the exact PyPI package link or the repo contents so I can check what the package installs and whether it reads the declared config paths.
Capability Assessment
Purpose & Capability
Name/description claim local MLX-powered inference on Apple Silicon; SKILL.md contains commands and examples that align with that (pip install ollama-herd, run herd/herd-node, curl local API endpoints). Declared required binaries (curl or wget) and optional python3/pip are proportional. Minor mismatch: SKILL metadata lists configPaths under ~/.fleet-manager, which suggests the skill expects to read or manage local fleet state; that is not clearly described in the prose but is plausible for a fleet/monitoring tool.
Instruction Scope
SKILL.md instructs installing a PyPI package (pip install ollama-herd) and running services (herd, herd-node), plus using 'uv tool install' for image backends. The instructions do not ask the agent to read unrelated system files or external secrets, and API calls are to localhost. Concerns: (1) 'uv' is used but not declared in the metadata as a required binary — the agent may need to fetch/execute an additional tool; (2) the metadata lists configPaths (~/.fleet-manager/...) but the runtime examples don't show explicitly reading them; if the skill or installed package accesses those paths, they could contain local telemetry or logs. No explicit instructions to exfiltrate data, but installing packages and running daemons will give code disk & network access under the user account.
Install Mechanism
This is an instruction-only skill (no install spec in registry). It tells the user/agent to pip install ollama-herd from PyPI and to run herd/herd-node. Pip installing a third-party package is a common pattern but introduces risk because code will be downloaded and executed locally. There are no downloads from obscure URLs in the SKILL.md. The 'uv tool install' step implies additional external tooling; the source and trust model for 'uv' are not documented here.
Credentials
The skill declares no required environment variables or credentials — consistent with local-only operation. The examples use localhost endpoints and a placeholder api_key 'not-needed'. No requests for unrelated cloud credentials or secrets appear in the SKILL.md. Declared configPaths may give access to local fleet logs/state, which is reasonable for a monitoring/orchestration tool but worth noting.
Persistence & Privilege
always is false and model invocation is allowed (default). The skill asks the user to install/run local services (herd, herd-node) which will persist as processes/files under the user account; this is expected for an orchestration/daemon tool. The skill does not request to modify other skills or global agent configuration.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install mlx-apple-silicon-mlx - After installation, invoke the skill by name or use
/mlx-apple-silicon-mlx - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.1
Cross-platform support: macOS, Linux, and Windows. Updated OS metadata, descriptions, and hardware recommendations.
v1.0.0
Initial release: MLX-powered local AI fleet for Apple Silicon.
- Run LLMs, Stable Diffusion, speech-to-text, and embeddings natively using Apple's MLX framework.
- Unified fleet router coordinates inference and generation across multiple Mac devices (Mac Studio, Mac Mini, MacBook Pro).
- Ollama (LLMs/embeddings), mflux (Flux image gen), DiffusionKit (SD3), and Qwen3-ASR (transcription) all use MLX backend.
- Metal-accelerated, Apple-optimized: no PyTorch, no CUDA, fully local, no external calls.
- Includes install/setup instructions, API usage examples, fleet monitoring, and guardrails for privacy and safety.
Metadata
Frequently Asked Questions
What is Mlx Apple Silicon Mlx?
MLX-powered local AI — run LLMs, Stable Diffusion, speech-to-text, and embeddings natively on Apple Silicon via MLX. Ollama uses MLX for LLM inference, mflux... It is an AI Agent Skill for Claude Code / OpenClaw, with 124 downloads so far.
How do I install Mlx Apple Silicon Mlx?
Run "/install mlx-apple-silicon-mlx" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Mlx Apple Silicon Mlx free?
Yes, Mlx Apple Silicon Mlx is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Mlx Apple Silicon Mlx support?
Mlx Apple Silicon Mlx is cross-platform and runs anywhere OpenClaw / Claude Code is available (darwin).
Who created Mlx Apple Silicon Mlx?
It is built and maintained by Twin Geeks (@twinsgeeks); the current version is v1.0.1.
More Skills