A turnkey AI platform that we preconfigure and tailor to your network. You plug it in and it is reachable on your local network. From that moment your team writes texts, summarises documents and builds its own knowledge.
The starting point
For harmless tasks a cloud service is convenient. As soon as HR, customer or client data is involved, the calculation turns around.
Comparison
AI Cube
Both together works as well: confidential work stays on the Cube, and for uncritical tasks you additionally approve a cloud model. You decide what goes where. Detailed comparison on the blog
From day one
Sign in through the browser and start. No third-party accounts, no rollout phase, no training week, nothing to install on the workstations.
Chat, workspaces, your own files and knowledge bases in one window. The same handling as the well-known AI services, except it runs in your own building. The interface is Open WebUI, an open-source platform for self-hosted AI, in your language and in the browser.
Open WebUI brings these features, and they run entirely on the Cube.
RAG and company knowledge
Manuals, minutes, policies and contracts are prepared for internal AI search. Staff ask in plain language and get answers with traceable sources.
Usable from delivery
Extension by WZ-IT
If content should not be maintained by hand, we connect existing repositories and business systems: with synchronisation, permissions, text recognition, metadata and quality tests. A separate project after go-live.
Connect your own knowledge sources · See internal AI assistants
Sovereignty
Local AI is not automatically sovereign AI. These four points make the difference, and every one of them can be verified.
No connection to external AI providers out of the box and no telemetry to us. Prompts, documents and answers stay in the building.
No vendor can switch off your access, no licence expires, no price list changes at the end of the quarter.
The device is yours. You can keep running the AI Cube without us at any time; the handover documentation is in place.
An encrypted channel through our VPN. On request it stays off and is opened only for a maintenance window.
Professional secrecy: Section 203
Wherever personal data is processed, the question is not whether the AI writes well but where the data goes. Companies with staff and customer records, municipalities with citizen data and confidentiality professionals with client and patient records all face the same question. The AI Cube answers it: the data goes nowhere, it stays here.
Tax advisers, lawyers, auditors, doctors and psychotherapists are additionally bound by professional secrecy. A data processing agreement alone does not resolve it: anyone involved as an assistant must be obliged in text form and instructed about criminal liability. When we look after the platform under AI Cube Care, we conclude that additional agreement on request.
The legal assessment of your use case stays with you and your advisers. We supply the technical and contractual basis.
Local AI for confidentiality professionals · Section 203 managed cloud
In their guidance on AI and data protection, the German data protection authorities describe technically closed systems as preferable from a data protection perspective.
For your IT team
What your IT team wants to know before the device joins the network. A device the size of a book, no server room, no cooling.
Unified memory, shared between CPU and GPU. This figure decides which models run and how many documents stay in context at once.
Grace Blackwell, CPU and GPU on one chip
Arm processor
2 TB and 4 TB as an upgrade
network port
under load, 240 watt power supply
quiet, a normal wall socket
Linux, hardened, automatic security updates in the maintenance window
vLLM as the inference engine, tuned to the hardware
Open WebUI: chat, knowledge bases, assistants, groups and roles in the browser
Open-weight models of your choice, your own models can be installed
Embedding model for knowledge bases and source references
OpenAI-compatible endpoints prepared, disabled out of the box
HTTPS on the local network with its own certificate, a fixed address in your network
On the device. No inbound port forwarding required
VPN agent (NetBird), encrypted, you can switch it off at any time
Configuration backup to your NAS, monitored
Local accounts with groups and roles; your directory as an extension
Documentation and root access stay with you
Up to 10 parallel chats per AI Cube: with the standard configuration one AI Cube generates up to ten answers at the same time. The figure describes running answers, not the number of user accounts: further staff have their own accounts and use the system at different times.
What matters is how many people generate an answer at the same moment, not how many users exist.
Long documents and extensive knowledge queries need more compute than short chat questions.
When it gets more, a second Cube joins, or a custom-built system with several GPUs.
Delivered with the power supply, a network cable and the handover documentation. Two-year manufacturer warranty. Designed for teams of five to twenty people working with confidential data: law firms, tax advisers, medical practices, engineering offices, municipalities and mid-sized companies.
What you get
Once for the device and its setup, monthly for maintenance. No token billing, no user licences, no price list that changes at the end of the quarter.
One-off
Available on top
Ongoing maintenance
A device that runs AI needs care. Operating system, drivers, inference engine, interface and models receive updates continuously. We handle that, so your team does not face an error message on Monday morning.
Care starts with the device and runs through the first twelve months. After that you decide month by month whether it continues; the AI Cube keeps running either way, then without updates, monitoring and support.
If you need a faster response, up to 24/7: higher service levels from €99.90 excl. VAT / month. View the service levels
How it works
From the first message to the first answer in the browser it takes four steps. You order after the conversation, not in the form.
You tell us the use case and roughly how many users. We clarify storage, network and operating scope and confirm the configuration.
We set the device up for your network, select and test the language models on the hardware, harden and document the system.
In North Rhine-Westphalia we set up on site and brief your team. Elsewhere we ship to Germany, Austria and Switzerland and brief you remotely. In the box: power supply, network cable and handover documentation.
Your team signs in through the browser. From then on AI Cube Care keeps the system up to date and monitored.
Build on later
The AI Cube is the beginning of your own AI and knowledge platform. Because open software sits underneath, you can build on it as your needs grow. The three extensions customers ask for most:
Active Directory, Entra ID or another identity provider. Your people sign in with the account they already have, with groups and roles from your directory.
The AI works where the work happens: in the ticket system, the CRM, the case management. Through the Cube's open interface, workflows can be connected without switching platforms.
Confidential work stays on the local model. For uncritical tasks you can additionally approve models from German providers or the well-known large models.
When you outgrow the Cube we build the next system: multiple GPUs, rackmount, more memory, matched to your workload. We stay your single point of contact, from the device to operations.
View custom systemsHow fast does each model answer on this hardware? The simulator shows it from measured values.
Open the token simulatorSystem houses and consultants can resell the AI Cube. Terms on request.
View the reseller programmeThese extensions are separate projects and not part of the purchase price. They do not hold up the start: the platform is working from the day of delivery, and you build on it afterwards at your own pace.
Role in the overall system
This page covers hardware, the AI platform and ongoing operations. Process assessment beforehand, knowledge integration and continuous quality control remain separate modules you can add as needed.
Clarify a recurring process, its data and sensible automation boundaries before hardware is in place.
View service Make knowledge usableConnect one priority knowledge source with citations, permissions and subject-matter test questions.
View service Control continuouslyContinuously evaluate traces, answer quality, latency and changes in production AI applications.
View serviceThree relevant adjacent paths instead of a long list of further products.
View the overall systemAnswers to the most important questions
Knowledge base
Reviews & projects
From local AI integration to architecture, data sovereignty and ongoing operations.
WZ-IT moved our studio infrastructure from decentralised individual devices to a central platform: every site is securely connected via VPN, new devices are onboarded automatically and an entire site is provisioned from a template, without manual steps on location. What impressed me most is the breadth and depth of their knowledge: Timo and Robin are not a typical IT provider who sets up a server and leaves. The two of them think their way into highly complex infrastructure and software topics, work through every requirement we put in front of them, and build networking, provisioning and operations so that everything fits together in the end. WZ-IT is an excellent partner for complex software, network and architecture projects.
Steve KirchnerManaging Director, nextGYM GmbH
International
WZ-IT designs, develops and operates infrastructure and software for clients in Germany and internationally. We deliver projects remotely and continue supporting them in ongoing operations after go-live.
Enquiry
Tell us briefly how you want to use local AI and how many people will work with it. We will review the configuration, integration and suitable operating scope and come back to you with a recommendation.
No risk: worst case, you leave with a clearer understanding of your project than before.
WZ-IT's advice on our Azure migration was technically sound and completely non-binding right from the intro call - we took away a great deal.
Jakob ÖschlbergerInno7 GmbH
WZ-IT's advice on our Azure migration was technically sound and completely non-binding right from the intro call - we took away a great deal.
