MLX

Apple’s array framework. On an M-series Mac it is the fastest way to run a model locally.

mlx-lm is Apple’s array framework pointed at language models. On an M-series Mac it is the fastest local path, because it uses unified memory instead of fighting for a discrete GPU.

Apple Silicon only. `mlx_lm.server --model mlx-community/Qwen3-8B-4bit` serves OpenAI-compat on localhost:8080/v1. Context is whatever your unified memory holds.

Max context: Whatever your unified memory allows
Free tier: Unlimited and private
Requirement: Nothing — no account

Endpoint

Base URLhttp://localhost:8080/v1

Rate limitNone — it is your machine

OpenAI-compatible
Native SDK: mlx-lm
text
vision

Code examples

npm install openai

import OpenAI from 'openai'

const client = new OpenAI({
  baseURL: 'http://localhost:8080/v1',
  apiKey: 'not-needed'
})

const response = await client.chat.completions.create({
  model: 'mlx-community/Qwen3-8B-4bit',
  messages: [{ role: 'user', content: 'Explain closures in one paragraph.' }]
})

console.log(response.choices[0].message.content)

pip install openai

from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8080/v1",
    api_key="not-needed",
)

response = client.chat.completions.create(
    model="mlx-community/Qwen3-8B-4bit",
    messages=[{"role": "user", "content": "Explain closures in one paragraph."}],
)

print(response.choices[0].message.content)

curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mlx-community/Qwen3-8B-4bit",
    "messages": [{"role": "user", "content": "Explain closures in one paragraph."}]
  }'

pip install mlx-lm

from mlx_lm import load, generate

model, tokenizer = load("mlx-community/Qwen3-8B-4bit")

prompt = tokenizer.apply_chat_template(
    [{"role": "user", "content": "Explain closures in one paragraph."}],
    add_generation_prompt=True,
)

print(generate(model, tokenizer, prompt=prompt, max_tokens=512))

Worth knowing

Apple Silicon only.
Start the server first: `mlx_lm.server --model mlx-community/Qwen3-8B-4bit`.

Models you get for free3

Qwen 3qwen3
Qwen CoderQwen2.5-Coder-7B
GPT-OSS 120Bgpt-oss-120b