SpecsM4 Pro · 64GB · 2TB
Unified memory for models and build workloads. Also available: M4 · 16GB · 256GB.
Run MLX inference
on your own dedicated Cloud Mac
Rent by the day, with the entire machine reserved for your workloads. Connect over SSH and run MLX on Apple Silicon, plus Xcode builds and tests on the same node.
Dedicated physical machine, not a VM · Daily, weekly, monthly, or quarterly billing · Prices in USDStarting daily rateFrom $21.6/day
VMR Lab M4; pay in advance by the week, month, or quarter.
Inference servermlx_lm.server :8080
Listens on the node only; connect through an SSH tunnel.
VMRentalOrder setup / 01Choose a physical node
Give your experiment a Mac of its own
VMR Lab ProM4 Pro · 64GB · 2TB
Tokyo · JPSSH / VNC · Dedicated machine
Your configurationTokyo / M4 ProPrepaid daily · USD$60.9
SG JP KR HK US-W · Dedicated physical machine
9:41●●●
Node statusTokyo · JPPhysical node running- Chip
- M4 Pro
- Memory
- 64GB
- Remote access
- SSH / VNC
localhost:8080After you run ssh user@node your workloads are the only ones running on this Mac mini.
MLX / MODEL LOAD
$ pip install mlx-lm
$ mlx_lm.server --port 8080
loading weights → unified memory
server ready · 127.0.0.1:8080Model weights / KV cache / system tasks
Mac for AI
Rent a whole Mac for as long as your experiment needs
Apple Silicon uses unified memory, which CPU and GPU can access from the same pool. Choose VMR Lab Pro for 64GB of unified memory for model weights, caches, and system tasks. Available capacity depends on the processes running at the same time.
First, install mlx-lmon the node and configure the inference server to listen only on localhost. Then use an SSH tunnel to forward the port to your computer. When your experiment is over, you don't have to keep paying for a machine you bought just for the job.
- 01
pip install mlx-lm - 02
mlx_lm.server --port 8080 - 03
ssh -L 8080:127.0.0.1:8080 user@node
Hardware specs
What's on the label is what's inside
Both options are full, dedicated physical Mac minis—not shared VMs. Check unified memory and local storage before ordering. Add extra SSD storage if you need more space.
VMR LAB / MODEL PLATEDEDICATED PHYSICAL MAC MINI
| Specs | VMR Lab M4 M4 · 16GB · 256GB |
VMR Lab Pro M4 Pro · 64GB · 2TB |
|---|---|---|
| Chip | Apple M4 | Apple M4 Pro |
| Unified memory | 16GB, suited to everyday Xcode builds and smaller quantized models | 64GB, suited to larger context windows or multi-model experiments |
| Local storage | 256GB SSD | 2TB SSD |
| Node connectivity | Remote access from your selected location; actual latency depends on your connection | Remote access from your selected location; actual latency depends on your connection |
| Remote access | SSH / VNC | SSH / VNC |
| Resource allocation | Dedicated machine, not a VM | Dedicated machine, not a VM |
SG · JP · KR · HK · US-WPrices in USD
Published pricing
Two models, available by the day
Choose enough memory for your workload, then pick a daily, weekly, monthly, or quarterly term. Prices below are prepaid in USD; optional add-ons cost extra.
Optional add-ons are billed separately for your chosen term: +1TB SSD from $2.2/day, +2TB SSD from $4.4/day, and Thunderbolt 5 daisy-chaining from $1.3/day per machine. Availability at checkout is confirmed by the console in real time.
One node · Two workflows
Build with xcodebuild by day, run inference by night
Use SSH for Xcode builds and simulator tests during the day, checking compatibility across iOS and macOS versions. Once the build is done, use fastlane to prepare your TestFlight release.
At night, stop the build processes and let mlx_lm.server run persistently. Your projects, dependencies, and model files stay on the same dedicated physical machine, so there's no need to keep moving your environment between tasks. Before running workloads in parallel, make sure you have enough memory and disk space.
NODE LOG / JPA day on one physical nodeWorkflow illustration, not a usage log
xcodebuild -scheme App buildBuildxcodebuild testTestfastlane betaDistributemlx_lm.server --port 8080Inference
Physical nodes
Five locations. Choose the best route for you.
Both catalog models are available in Singapore, Tokyo, Seoul, Hong Kong, and US West. Latency ranges below are estimates, not a guarantee of your actual connection. Check real-time availability in the console.
Singapore
Routes to Southeast Asia
Est. 20–50 msM4 / M4 ProTokyo, Japan
Routes to East Asia
Est. 30–60 msM4 / M4 ProSeoul, South Korea
Routes to Northeast Asia
Est. 30–50 msM4 / M4 ProHong Kong
Routes to southern China and Southeast Asia
Est. 20–40 msM4 / M4 ProUS West
Routes to the US West Coast
Est. 25–35 msM4 / M4 ProWorkflows
Four ways to work on one dedicated machine
Check your toolchain and resource needs before choosing a model. All four use cases start with remote access; memory, storage, and runtime vary by workload.
01 / Indie developers
From build to release
Build Xcode projects remotely, run tests, then use fastlane to prepare your TestFlight release.
fastlane betaView workflow →
02 / CI teamsAlways-on build node
Run a self-hosted runner on a dedicated Mac, triggered by a PR with xcodebuild test.
xcodebuild testView workflow →
03 / AI experimentsKeep models on the node
Choose M4 Pro with 64GB to keep quantized models in memory; your team can access the inference service through an SSH tunnel.
mlx_lm.serverView workflow →
04 / Audio and videoKeep media and rendering together
Transcode and render on a Cloud Mac. Need more room for your media? Add a +2TB SSD.
+2TB SSDView workflow →
Remote work, visualized
Close your laptop. Your workloads keep running.
Create an SSH tunnel from your laptop to access services on the node. For a graphical interface, connect first, then use remote desktop.
LOCAL → SSH TUNNEL → MAClocalhost:8080
REMOTE SESSION / SCENARIO
Keep the connection in your workspace
Call Cloud Mac model services from your laptop while builds and inference run on the dedicated physical node. Step away from your desk and your tasks and files stay on the machine you rented.
View the remote workflowBefore you order
Four questions to help you decide
Is it a physical Mac or a VM?
You rent a whole physical Mac mini. Its CPU, GPU, and unified memory are not shared with other customers.
Which model should I choose for MLX?
Check the memory needs of your quantized model, context cache, and parallel processes. For everyday builds or smaller models, start with M4 16GB. For more memory, choose M4 Pro 64GB.
How do I forward the inference port to my computer?
Configure the service to listen on 127.0.0.1:8080then run ssh -L 8080:127.0.0.1:8080 user@node to forward the port locally.
How do billing periods and payments work?
Prepay by the day, week, month, or quarter. Pay in USD with USDT-TRC20 or Visa / Mastercard / Amex through Stripe. Available payment options are confirmed by the backend.
Lab notes
Deployment guides and workflow notes
Reusable steps for connecting, building, and choosing a billing term.
DevOps & CI/CD
Test Idempotent Replay of an iOS Offline Write Queue on a Cloud Mac
Reproduce a committed write followed by a failed response, then verify that retries and app restarts cannot create a second record.
DevOps & CI/CDTest iOS haptics on a cloud Mac without trusting the simulator to feel
The simulator can inspect haptic event timing, not the sensation. Use this workflow to separate repeatable CI checks from real-device acceptance.
DevOps & CI/CDTest file watchers on a cloud Mac without counting callbacks as edits
A runnable directory-watching experiment for rapid writes, atomic saves and snapshot-based verification.
New posts are in the works. In the meantime, browse theblog.
Next / Configure your node
Choose a model, term, and location to get started
Choose between two dedicated physical Mac mini models at five locations. Not sure how much memory or time you need? Compare the full pricing first, or configure your rental when you're ready.