Quickstart

#Run a model on your Mac

1. 1

   Install mlx-lm

   On a Mac with Apple silicon: pip install mlx-lm.

2. 2

   Start a model

   mlx_lm.server --model mlx-community/Mistral-7B-Instruct-v0.3-4bit downloads the model the first time and serves it on port 8080 of localhost. mlx.app/compare shows which models fit your Mac's memory.

3. 3

   Call it

   POST OpenAI-style chat JSON to http://127.0.0.1:8080/v1/chat/completions. The fields and options are in Local AI quickstart.

#Read the network

The network API is public JSON. No key, no sign-up.

curl https://mlx.app/api/stats

const res = await fetch("https://mlx.app/api/stats");
const stats = await res.json();
console.log(stats.head.height, stats.stats.nodes);

import requests

stats = requests.get("https://mlx.app/api/stats").json()
print(stats["head"]["height"], stats["stats"]["nodes"])

let url = URL(string: "https://mlx.app/api/stats")!
let (data, _) = try await URLSession.shared.data(from: url)
let stats = try JSONSerialization.jsonObject(with: data) as! [String: Any]
print(stats["head"] ?? "")

From here, blocks, transactions, wallets and the compute rewards ledger work the same way. Amounts are strings of base units with 6 decimals.

#Put a Mac to work

1. 1

   Install MLX Node

   Download it from mlx.app/download and open it: it installs itself and appears in the menu bar.

2. 2

   Link it to a wallet

   MLX Node opens MLX Wallet and asks you to create or sign in to a wallet; the link is a transaction signed by your wallet and by the Mac's Secure Enclave key. A wallet can link up to 32 devices.

3. 3

   Start it

   New installs run when you press Start. Set an idle time in Settings to have MLX Node work only while you're away. A Mac on a free seat mines up to 1 MLX a day; dedicated seats mine more.