Done-for-you · On site in Belgium and the Netherlands
DFY Local AI Installation: your own AI without filters or credits
With a Local AI installation you get a complete AI environment that runs inside your own company. We supply the hardware and the software and install everything on site. Then we follow up for 2 weeks.
Prefer to call? +32 487 24 75 52
- From €20,000
- Mac Studio M5 Ultra with 256GB unified memory
- On-site installation + 2 weeks of follow-up
What is a Local AI installation?
A Local AI installation is an AI system that runs entirely on your own hardware. Not in someone else’s cloud, but in your office. You pay no subscription and no per-token costs.
We do everything for you. We supply a Mac Studio M5 Ultra with 256GB unified memory. On it we install open-source models that run fully locally, including abliterated variants. These are uncensored, “jailbroken” versions without built-in refusals.
The models work together in a multi-agent setup. A reasoning model thinks and plans. A fast executing model carries out the tasks. That gives you private AI that gets real work done, within the walls of your company.
Why local AI?
-
No filters, no refusals
The AI does what you ask. No lectures, no “I can’t help you with that”.
-
No credits or subscriptions
No monthly licences and no per-token costs. You use the AI as much as you like.
-
No constant permission clicks
No endless pop-ups to approve every step. The agents keep working.
-
Your data stays in-house
The system runs on an isolated device. Your data never leaves your company. That is safer and private.
-
Agents that work on their own
Agents can create accounts, log in, operate tools and handle payments on their own.
-
Payments with a fixed budget
Our recommendation for payments. Give the agent its own card with a limited budget, for example a Revolut virtual card with a spending limit.
Why a Mac Studio M5 Ultra for local AI?
In our view, the Mac Studio with M5 Ultra is the most powerful local AI rig for small and medium-sized businesses. We only install the most powerful chip Apple sells today: the M5 Ultra with a 36-core CPU, an 80-core GPU and a 32-core Neural Engine, with 256GB unified memory.
The reason lies in technology Apple developed itself. In a classic PC, video memory is separate from system memory, and an AI model has to fit into that smaller video memory. Apple silicon uses unified memory: the CPU, GPU and Neural Engine share one pool. That lets a model use almost the full 256GB.
Then there is speed. The M5 Ultra reaches a memory bandwidth of 1.2 TB/s. With large language models, that bandwidth largely decides how fast the AI answers. With MLX, Apple also has its own framework for running AI models on Apple silicon.
- Chip: Apple M5 Ultra
- CPU: 36 cores (12 super cores and 24 performance cores)
- GPU: 80 cores
- Memory: 256GB unified memory
- Memory bandwidth: 1.2 TB/s
Mac Studio M5 Ultra vs AMD and Nvidia
| Mac Studio M5 Ultra | AMD Ryzen AI Max+ 395 | Nvidia workstation | |
|---|---|---|---|
| Memory for AI models | 256GB unified memory | 128GB at most | 288GB with 3 x RTX PRO 6000 |
| Memory bandwidth | 1.2 TB/s | 256 GB/s | 1.8 TB/s per card |
| Two 70B models at Q8 at once | Yes | No | Yes |
| Price | from €20,000, all-in | lower | €50,000 to €100,000 or more |
An AMD system is cheaper, but it tops out at 128GB of memory and 256 GB/s. You cannot run a decent multi-agent setup on it with two large models at high precision. The Mac Studio can, with almost five times the memory bandwidth, so it is much faster too. For an equivalent Nvidia setup you quickly pay €50,000 to €100,000.
Which local LLMs do we install?
Unless you ask for something else, we install this multi-agent setup. It is the setup we currently install for clients.
-
Thinking agent
DeepSeek-R1-Distill-Llama-70B (Q8). Uses about 75GB of memory. This model reasons, plans and checks the work.
-
Coding agent
Llama-3.3-70B-Instruct (Q8). Uses about 75GB of memory. This model writes the code and carries out the tasks.
Why this combination?
- Fast and smart: the maximum for a Mac Studio with 256GB.
- Two models from different makers, DeepSeek and Meta: the best basis for a multi-agent workflow.
- Q8 is 8-bit: almost no quality loss compared with the full model.
- Both models can work with their full context window.
- Enough memory is left for macOS and the tasks on the device itself.
Naturally, we install the abliterated (“jailbroken”) variants of these models. That is one of the main reasons to buy your own AI rig: you hand work over completely to an AI agent that refuses nothing, keeps working and also handles payments, registrations and accounts.
What’s included
- Hardware: a Mac Studio M5 Ultra (36-core CPU, 80-core GPU) with 256GB unified memory
- Software: DeepSeek-R1-Distill-Llama-70B and Llama-3.3-70B-Instruct as a multi-agent setup, in abliterated variants. Other models on request.
- On-site installation: we come to you and get everything up and running
- Training: we show you and your team how to work with the system
- 2 weeks of follow-up: we stay available for questions and adjustments
How it works
-
Intake
We discuss which tasks you want to hand over to AI.
-
Setup
We order the Mac Studio and prepare the models. By default DeepSeek R1 as thinking agent and Llama 3.3 70B as coding agent, or other models if you prefer.
-
On-site installation
We install everything at your office, configure the agents and train your team.
-
Follow-up
For two weeks we answer questions and make adjustments.
What do you use it for?
-
Marketing and SEO automation
Let agents handle content, keyword lists and recurring marketing tasks.
-
Research
Search, summarise and compare sources, without sensitive questions leaving the building.
-
Internal knowledge
Let the AI work with your own documents and processes. Everything stays on your own AI server.
-
Coding
Write code, review it and build scripts with a local LLM that does not refuse.
-
Admin
Fill in forms, manage accounts and handle routine tasks in your tools.
Why have a local AI rig installed now?
-
The M5 Ultra has just arrived
The Mac Studio with M5 Ultra has been on sale since September 2026. We do not expect a better option before 2028. Buy now and you are set for years.
-
Delivery times are getting longer
Delivery times for the Mac Studio with M5 Ultra keep increasing. We suspect more and more companies are discovering this gap in the market.
-
Memory is getting more expensive
RAM is getting more expensive worldwide because of demand from AI data centres. We expect the price of the Mac Studio to rise soon.
In short: this is the perfect moment to have your local AI rig installed.
What does a Local AI installation cost?
One package, no subscription afterwards.
-
DFY Local AI Installation
From €20,000
from €20,000 (including all hardware, software and installation + 2 weeks of follow-up)
- Mac Studio M5 Ultra with 256GB unified memory
- Open-source models in a multi-agent setup
- On-site installation and training
- 2 weeks of follow-up
Frequently asked questions about Local AI
What does a Local AI installation cost?
The installation costs from €20,000 (including all hardware, software and installation + 2 weeks of follow-up).
Which models run on it?
By default we install two models in a multi-agent setup: DeepSeek-R1-Distill-Llama-70B as thinking agent and Llama-3.3-70B-Instruct as coding agent. Both at Q8 and in abliterated variants. Would you like other models, such as Qwen or GLM? That is possible on request.
Why a Mac Studio and not an Nvidia or AMD system?
The Mac Studio M5 Ultra combines 256GB unified memory with a memory bandwidth of 1.2 TB/s. An AMD system with the Ryzen AI Max+ 395 reaches 128GB and 256 GB/s at most. An equivalent Nvidia setup quickly costs €50,000 to €100,000.
Which Mac Studio do you install?
Only the most powerful chip Apple sells today: the M5 Ultra with a 36-core CPU, an 80-core GPU and 256GB unified memory.
Is this legal?
Yes. Running open-weight models locally is legal. What you let the AI do is your own responsibility. Lawful use is up to you.
Do we still need a subscription?
No. Everything runs locally. You pay no credits, no subscription and no per-token costs.
Where do you install?
On site in Belgium and the Netherlands. Elsewhere on request.
Related services
-
LLM link building
View LLM link buildingBecome visible in the answers of AI assistants and chatbots.
-
SEO consultant
View SEO consultantStrategic SEO advice from a specialist, direct and to the point.
-
Programmatic SEO
View Programmatic SEOBuild hundreds of pages at scale from data. Ideal to combine with AI agents.
Contact us ☕️
Talk to one of our consultants today and learn how to start leveraging your business.