Local AI set-up · on site, in Australia
Private AI that runs in your office
Open-weight models chosen for your work, running on hardware you own: a machine in your office, or a laptop for each person.
We’ll show you what local AI can do on your team’s real tasks before you spend anything on hardware. When you’re ready, we source and configure the hardware, install it at your office, and give your IT team a guide to keep it secure.
$ cluster status
4 nodes 512 GB for models 0 requests leaving the building
Why local
Your data never leaves the building
Local AI means the models run on machines you own, on your own network. Nothing your team types is sent to an AI company.
-
Your data stays yours
Prompts, documents and customer records stay on hardware you own. No third-party retention policy to vet for every new tool.
-
It can’t change under you
Hosted models are updated, repriced and retired on someone else’s schedule. Weights you own change only when you decide.
-
Costs you can plan
You pay once for the hardware, not per token. A busy month doesn’t turn into a surprise invoice.
-
You can check it
The exact models and settings you run are the ones we test, and they can be tested again after every change.
Choose your hardware
Pick a size. Grow it later.
Choose the hardware first: a cluster of NVIDIA DGX Sparks, a single Mac Studio, or a laptop for each person. It is the same set-up at a different size, and more memory holds larger models that handle longer, harder work.
For the office
| Size | DGX Spark cluster | Mac Studio | Good for |
|---|---|---|---|
| Starter | One DGX Spark, 128 GB | 96 GB | Useful day-to-day tasks: drafting, summarising and answering questions about your documents. |
| Max | Two DGX Sparks, 256 GB | 256 GB | Heavier everyday work. |
| Max+ | Three DGX Sparks, 384 GB | Not offered | Short, complex tasks. |
| Ultra | Four DGX Sparks, 512 GB | 512 GB | Complex tasks that run for many steps. |
For each person
| Size | MacBook Pro, M5 Max | RTX Spark laptop, Windows | Good for |
|---|---|---|---|
| Starter | 48 GB | Not offered | Smaller models for everyday drafting, summarising and questions. |
| Max | 64 GB | 64 GB | Mid-size models and longer documents. |
| Ultra | 128 GB | 128 GB | The largest models a laptop can run. |
Every set-up includes
- The hardware, sourced by Azul Labs and configured before it arrives
- A few open-weight models, picked for your tasks
- Local agents and the software your team uses to work with the models
- Set-up in person at your office
- A user guide, and a security guide for your IT team
One fixed price per size
Each size has a single price that covers the hardware and the set-up, quoted up front. Hardware prices move with suppliers, so we quote on the day.
Ask for a quoteHow it works
Six phases, from demo to hand-over
-
Before you buy
1. Discovery
Can local AI solve your problem? We talk through what you want it to do and demo it live, so you see what open-weight models can do today.
-
Before you buy
2. Evaluation
We capture day-to-day tasks from your team and run them on local models, so you see what it does on your work, not on someone else’s examples.
-
Before you buy
3. Choice
You choose the hardware and the size that fit, with a fixed price and a recommended set of models. You can move up as your use grows.
-
After you choose
4. Source
We source the hardware and configure it with the recommended models and local agents.
-
After you choose
5. Set-up
We come to your office and set it up ourselves: installed on your network and working for your team before we leave.
-
After you choose
6. User guide
Instructions for common issues and how to fix them, and how your IT team can secure the hardware within your network.
Security
Secured on your network, your way
Your IT team can secure it themselves, following a guide written for them to check against their own policies. Or we recommend a secure set-up for your network and walk IT through it.
-
Common issues
What can go wrong day to day, and how to fix it without calling us.
-
On your network
Where the hardware sits, and what it does and doesn’t need to reach.
-
Who can use it
How staff connect, and how IT controls who gets access.
-
Keeping it current
How to update the system and add new models safely.
Measured, not guessed
We test these models on hardware we run
We maintain Forcebench, an open benchmark of AI models on real Salesforce work. It runs open-weight models on our own hardware, including a cluster of GB10s, the chip inside a DGX Spark, and records the hardware, quantisation and engine behind each score. So the models we recommend are ones we have run ourselves.
-
272 tasks
Salesforce tasks in the benchmark.
-
15 local set-ups
Measured on GB10s, from one to four per set-up.
-
Published
Forcebench v0.2.0, 25 October 2026, at forcebench.ai.
FAQ
Questions people ask
How much does a local AI set-up cost?
Each size has one fixed price that covers the hardware and the set-up, quoted up front. Hardware prices move with suppliers, so email leo@azl.au for a current quote.
Should we choose DGX Sparks or a Mac Studio?
Both can grow: you add Sparks to a cluster, and Mac Studios can be wired together too. The difference is the software. DGX Sparks run NVIDIA’s CUDA stack, which most AI software is built and tested on first, and it is mature at serving many people and agents at the same time. A Mac Studio runs models on Apple’s Metal, which is newer for this work: it is quick and quiet for one or two people at a time, and slows sooner when many requests arrive together. We recommend one after seeing your tasks in the Evaluation phase.
Can you set up laptops instead of an office machine?
Yes. We set up MacBook Pros and RTX Spark laptops with local models and agents, one per person. A laptop can run its own models on the road and use the machine in your office for heavier work when it’s on your network.
What if local AI can’t do what we need?
Then we’ll tell you in Discovery, before you spend anything on hardware.
Start with a demo
Book Discovery and we’ll show you what local AI can do on your kind of work. If it can’t solve your problem, we’ll tell you before you spend anything on hardware.
leo@azl.au