design

mlx-vlm: Run Vision‑Language Models Locally on Apple Silicon

Leverage mlx‑vlm to perform inference and fine‑tune VLMs directly on your Mac with MLX—ideal for founders building AI‑driven visual products. 5.6k+ GitHub stars.

From the repo · “MLX-VLM is a package for inference and fine-tuning of Vision Language Models (VLMs) on your Mac using MLX.”

intermediate⏱ 30 minutes💵 Free
5,640 stars+19★ this week852 forksPythonGuide quality 9/10 · reviewed Oct 15Updated 10/21/2026100% free · open source
What it is

Run Apple Silicon-based vision language models locally on your Mac.

Real use case example

"A founder wants to develop a product that can detect and classify clothing items in images. They start by installing MLX-VLM and importing the model. Next, they integrate it with their app using Python code to create a working prototype. The outcome is a product that accurately identifies images of clothing items and provides insights for their e-commerce business."

Best insideClaude DesktopSelf-hosted
When to use it
  • •You're building an AI application that requires both image and text understanding
  • •You need to fine-tune a pre-trained vision-language model for your specific use case
  • •You prefer to develop and test AI models on your Mac rather than relying on cloud services
Ready-to-paste prompt
from mlx_vlm import VLM; model = VLM.from_pretrained('vlc-bert-base'); model.predict(image='path/to/image.jpg', text='This is a picture of a cat')
Saves to your device
Use with Claude
New

Skip the builder — one click puts this in Claude, Cursor, Antigravity and more.

⚡ Runs out of the box

A prompt/skill package — nothing to install beyond adding it to your AI tool.

Try it instantly — no install
Claude Code
mkdir -p ~/.claude/skills/mlx-vlm && curl -fsSL https://workflowstacks.com/api/skills/mlx-vlm/claude-skill -o ~/.claude/skills/mlx-vlm/SKILL.md
Open in another AI app

Opens the app with this repo with the prompt ready to go — no copy-paste needed.

Connect the whole catalog (MCP)
claude mcp add --transport http workflowstacks https://workflowstacks.com/api/mcp

Adds a WorkflowStacks connector to Claude Code: search and load any skill here by chatting.

How mlx-vlm: Run Vision‑Language Models Locally on Apple Silicon works
Codeflow
Free to inspect

Mlx-vlm: Run Vision‑Language Models Locally on Apple Silicon is a very large Python project (~280k lines across 1427 code files, plus 17 test files). You install it into your AI tool with one command; there is nothing to run yourself. Last commit this month, MIT license, has a test suite.

Size
Very large codebase
~280k lines · 1427 code files · days to read — use, don't read
Setup
Install as a skill / plugin
Add it to Claude Code (or your AI tool) with one command — 8 skills inside. Nothing to run yourself.
Runs on
Inside your AI tool
Helper scripts use Python
Python 100%
Where to start reading
  1. 1
    README.md
    Start here — what it does and how to install it
  2. 2
    mlx_vlm/__main__.py
    Where the program starts running
  3. 3
    mlx_vlm/__init__.py
    Inside mlx_vlm/ — the main logic begins here
  4. 4
    examples/dots_mocr_demo.ipynb
    A worked example — copy this to get going
What's in each folder
mlx_vlm/Core code — the actual logic1508 files
skills/Prompts, skills & agent definitions12 files
examples/Examples you can copy35 files
docs/Documentation12 files
.claude-plugin/Plugin manifest — what gets installed1 files
.github/CI / automation (GitHub Actions)4 files
READMEHas testsDocumentedCI checksExamples includedMIT licenseUpdated this month
Quick Actions
Details
Creator
Blaizzy
Language
Python
Category
design
Published
4/16/2024

Are you the creator of this tool? and earn 85% of every sale.

Show it off in your README:

Featured on WorkflowStacks
[![Featured on WorkflowStacks](https://workflowstacks.com/api/badge/mlx-vlm.svg)](https://workflowstacks.com/skills/mlx-vlm?utm_source=github&utm_medium=badge)
🔥 Hot this week, in your inbox
Skills like this, every Monday.

The five fastest-growing open-source AI skills, ranked by GitHub star growth. One email a week, unsubscribe anytime.