MLX INFERENCE · DEDICATED MAC MINI

Your MLX inference server,running on a dedicated Mac mini

Apple Silicon unified memory is shared by the CPU and GPU, so fully quantized 4-bit models load directly into memory. Launch an OpenAI-compatible endpoint with mlx_lm.server and start sending requests.

Every location runs on a real physical Mac mini, not a virtual machine. Choose Singapore, Tokyo, Seoul, or Hong Kong. Rent by the day and connect over SSH about 4 minutes after ordering.

Deploy from $19.4/day Compare the 3 models

Entry model: XOps 16 · M4 · 16GB · 256GB · $19.4/day · $96.8/month

mlx-server · Tokyo · XOps 24 M4 · 24GB · 512GB
Inference session · VerifiedSingle tenant · Physical node · Not a VM

Model weights loaded4.52 / 4.52 GB · 100%

8B Instruct · 4-bit quantization · safetensors
Session unlocked: dedicated physical machine ready READY
Unified memory used6.1 / 24 GBIllustrative
Tokens per second38.6 tok/sIllustrative

SSH tunnelssh -N -L 8080:127.0.0.1:8080 xops@tyo-24

EndpointPOST 127.0.0.1:8080/v1/chat/completions

  • Dedicated physical machineNot a VM; the entire machine is assigned to one tenant

  • 4 Asia-Pacific locationsSingapore · Tokyo · Seoul · Hong Kong

  • USDT-TRC20 and Visa · Mastercard · AmexCard payments processed by Stripe; billed in USD

  • About 4 minutes from order to readyTrack actual progress in your console order

MLX inference · mlx-lm

Run models in unified memory, with endpoints private to your team

mlx-lm loads quantized weights and runs inference on Apple Silicon. Start an OpenAI-compatible API with mlx_lm.server, bind it to localhost, then forward access through an SSH tunnel to team members who need it. This avoids exposing the inference port directly to the public internet.

Model files, dependencies, and services stay on the same dedicated physical machine. Use launchd to manage persistent services. When adjusting context length or concurrency, check the combined memory used by model weights, caches, and system processes—unified memory capacity is not all available for weights.

pip install mlx-lm
mlx_lm.server --model ./models/8b-instruct-4bit --port 8080
ssh -N -L 8080:127.0.0.1:8080 xops@tyo-24
Read the MLX deployment guide
iOS / macOS builds · CI

Your builds run on your machine, with no neighbors competing for resources

Install and switch Xcode versions to suit each project; use fastlane for signing, archiving, and distribution. Register the machine as a GitHub Actions or GitLab self-hosted runner to run jobs directly on this Mac mini, keeping DerivedData and dependency caches local.

Add certificates and provisioning profiles to a Keychain you manage, then run the runner under a controlled account. For parallel jobs, set concurrency based on available CPU, unified memory, and disk space—not as if a single machine were an infinitely scalable shared queue.

  • XcodeSwitch toolchains per project and keep builds reproducible.
  • fastlaneAutomate signing, archiving, and test distribution.
  • RunnerConnect to GitHub Actions / GitLab self-hosted pipelines.
Read the CI setup guide
Locations and provisioning

Dedicated nodes in four cities, rented by the day

Choose from three listed models in Singapore, Tokyo, Seoul, or Hong Kong. When selecting a location, consider where your team works and where your model files are stored. Latency tests are a useful reference, but measure the full connection from your own network. Check the console for live availability.

  • 4 Asia-PacificAsia-Pacific locations
  • 3 planslisted models
  • $19.4starting price / day
  • ≈4 minfrom order to connection
SINSingaporeSG · 1.35°N
TYOTokyoJP · 35.68°N
SELSeoulKR · 37.57°N
HKGHong KongHK · 22.32°N
From payment to connection

About 4 minutes from order to first SSH login

These are estimated provisioning times, not a fixed guarantee for every order. After payment is confirmed, the console shows allocation and setup progress. Check the console for live availability and completion status.

≈ 4.0 minChoose and pay → Allocate node → Initialize → First login
  1. 1

    Choose a model and location, then pay

    Confirm the model, rental period, and location, then complete payment in USD using an available payment gateway.

  2. 2

    Assign a dedicated physical node

    A physical Mac mini matching your selected model is assigned to your order.

  3. 3

    Initialize the system and issue credentials

    After provisioning checks, connection details appear in your console order.

  4. 4

    First login with SSH / VNC

    Verify your SSH connection first. If you need a graphical desktop, use VNC through an SSH tunnel.

Models and pricing · USD

Three models, rented by the day

All prices are in USD. Daily and monthly rates are shown below; see the full plans page for weekly and quarterly rates and storage add-ons. All models are available in Singapore, Tokyo, Seoul, and Hong Kong. Check the console for live availability.

16GB

XOps 16

Small quantized model experiments · Single-project CI

M4 · 16GB unified memory · 256GB SSD

$19.4/day$96.8/month
Deploy XOps 16
24GB

XOps 24

Run 7B–8B quantized models · Team CI

M4 · 24GB unified memory · 512GB SSD

$40.8/day$203.8/month
Deploy XOps 24
64GB

XOps Pro 64

Larger models or parallel workloads · Heavy builds

M4 Pro · 64GB unified memory · 2TB SSD

$61.2/day$305.8/month
Deploy XOps Pro 64
Choose a daily, weekly, monthly, or quarterly rental period. Renew or release your machine in the console.See weekly, monthly, and quarterly rates
What dedicated means

A whole machine, flexible terms, two remote access options

Dedicated physical machine, not a VM

The entire machine’s CPU, GPU, and unified memory are assigned to one tenant. Inference and build jobs don’t compete with other users on the same machine. Performance still depends on your models, projects, and concurrency settings.

Daily, weekly, monthly, or quarterly billing

Rent by the day for short-term testing, by the week for a release push, or compare monthly and quarterly rates for ongoing work. Before choosing a term, factor in actual usage time, model sync costs, and when your jobs will finish.

SSH and VNC access

Use SSH to write code and manage services; use VNC remote desktop when you need Xcode’s graphical interface. We recommend connecting to the desktop over an SSH tunnel and testing input latency on your local network first.

Workflow examples

Three common ways to use it

These typical scenarios can help you decide which workloads to run on a dedicated Mac.

“Put an 8B quantized model on a Tokyo node and let the team call its API through an SSH tunnel—without tying up their everyday laptops for inference.”
MLX inference · Internal team endpoint
“Run CI on a dedicated Mac mini and keep dependencies and build caches local. Push code and send jobs straight to your own build queue.”
Mobile development · Self-hosted runner
“Set up a release environment by the week. Archive, sign, and validate on the same machine, then export any artifacts you need before the job ends.”
iOS project · Release sprint
Before you order

Four questions about models, payments, and data

Browse more FAQs

Why choose a dedicated physical machine?

Each rental gives you one physical Mac mini, not a virtual machine sharing resources with other tenants. Your model weights, build caches, and jobs all run on the machine assigned to you.

What can I do with 16GB?

Start with small 4-bit quantized models, then measure available memory after accounting for weights, context cache, and concurrency. For running 7B–8B models, compare the 24GB model; for larger or parallel models, see the 64GB model.

Which payment methods are supported?

Pay with USDT-TRC20 or Visa, Mastercard, and Amex processed by Stripe. All payments are in USD. Check the console at checkout for currently available payment gateways.

What should I prepare before ending a rental?

Sync your model weights, project files, certificate backups, and any build artifacts you need to keep. The entire disk is erased when the machine is released, so don’t rely on files on the node as your only copy.

Technical articles

For developers working on cloud Macs

All articles
DevOps & CI/CD

Test iOS IME Composition Before Validating a Form

Handle marked text before running field validation, then test both the validation contract and the real keyboard interaction.

 AIDevelopment

Test iOS image thumbnail downsampling on a cloud Mac

Set a pixel limit when ImageIO creates iOS thumbnails, then test the output against fixed image fixtures. Learn what size assertions cannot prove about peak memory.

 DevOps & CI/CD

Test iOS Rotation Without Losing Draft State

Run one UI test across portrait and landscape to catch lost draft text, obstructed controls, and failed saves.
Next · Choose your setup

Choose a Mac mini and get your inference server running today

Start with XOps 16 at $19.4/day, then choose a model based on your inference and build workloads. Confirm the rental term, location, and live availability before paying.

Deploy nowChoose a model