Unlock knowledge locally
Documents and internal knowledge can be made accessible through chat and RAG applications.
On-premise. Preconfigured. Ready to use.
The plug-and-play AI solution for your company.
Preconfigured on-premise AI for teams, knowledge work and internal assistants – in your network and on your hardware.
stark AI Box
locally ready
01
Inside your network
Data and models remain in infrastructure under your control.
02
Available offline
Local AI applications can operate without cloud access.
03
Prepared for teams
Accounts, permissions, APIs and requested models are configured initially.
04
6 months of updates
The stated base package includes six months of software updates.
AI under your control
A system that goes beyond the hardware.
The AI Box combines capable workstation hardware with a ready-to-use software stack for secure, repeatable AI work in your company.
Control access cleanly
Team accounts, keys and permissions enable controlled use instead of shared personal accounts.
Integrate applications
Local APIs connect models with internal assistants and existing workflows.
Two performance tiers
The right box for your team.
The guide values refer to concurrently active users. Model, context length and specific tasks affect actual utilisation.
For everyday AI work
Box 32
from €8,900 netFor chat, email, knowledge work and compact internal assistants in everyday team use.
approx. 4–8 concurrently active users
Example: Qwen2.5 14B Instruct
GPUNVIDIA GeForce RTX 5090 · 32 GB VRAM
CPUAMD Ryzen 9 7900
RAM64 GB
Data1 TB NVMe
System2 × 512 GB SSD · RAID 1
Team chat
Text and email work
Local knowledge search
For larger models and RAG
Box 48
from €13,900 netFor larger models, broader knowledge systems, larger teams and sustained server operation.
approx. 6–10 concurrently active users
Example: Qwen2.5 72B Instruct, quantised
GPUNVIDIA RTX PRO 5000 Blackwell · 48 GB ECC-VRAM
CPUAMD EPYC 9654P
RAM64 GB
Data1 TB NVMe
System2 × 480 GB Enterprise SSD · RAID 1
Larger language models
RAG and knowledge systems
Sustained team use
Included in the base package
Initial setup of team accounts
Setup of the requested models
Six months of software updates
Already configured
From the interface to governance.
Employees get an approachable chat interface. IT and accountable teams manage models, access, keys and usage rules through a local gateway.
Browser-based access
Use approved AI capabilities from the workplace without specialist software.
Team and permission management
Manage accounts and permissions transparently for groups and use cases.
APIs in the company network
Connect internal systems with approved models through controlled endpoints.
Monitoring and logs
Review operations, usage and technical events in one central place.
Local AI assistant
intranet://assistant
Summarise our service policy for the project team.
The policy prioritises documented handovers, clear accountability and approval before external changes.
3 internal sources
Models
04
Teams
06
Status
Local
Role-based access
API keys
Usage logs
One box, two perspectives
Simple for employees. Controllable for IT.
01
Employees and knowledge work
A familiar interface for writing, summarising, researching and using internal assistants.
Central browser access
Approved models
Shared knowledge sources
02
IT and the company
A locally controlled entry point for models, integrations and transparent access rules.
Company network
Permissions and keys
Monitoring and governance
Introduction
From your use case to an operational AI Box.
The final configuration depends on models, user count, data sources and your existing infrastructure.
- 01
Clarify requirements
We frame tasks, user groups, data sources and integration needs.
- 02
Configure the box
Hardware and software are prepared for the agreed operating scope.
- 03
Set up and hand over
Accounts and models are configured initially and operations are handed over together.
Frequently asked questions
What to know before making an inquiry.
Can the AI Box be used without internet access?
Yes. The local AI applications are designed to run inside the company network without cloud access.
How many people can work at the same time?
As a guide, Box 32 is intended for about 4–8 and Box 48 for about 6–10 concurrently active users. Actual capacity depends on the model and tasks.
Can other models be used?
Requested models are configured initially. We assess which models run well on the selected hardware based on the use case.
What happens after the six months of updates?
The flyer does not define services beyond the included update period. We clarify any further scope transparently in the individual proposal.
Are the listed prices binding?
No. They are net starting prices and guide values. Configuration, components and effort can change the final proposal price.
Non-binding product inquiry
Which AI Box fits your use case?
Tell us briefly about your team, tasks and intended integration. We will get back to you with sensible next steps.
team@stark-ai.de
Handled confidentially