Step 01
1. Plug in LionBox
Connect power and Ethernet. The mobile app finds the nearby box and starts guided setup.
From first setup
How LionBox runs private AI on your office network
LionBox runs the AI model, your document index and chat history on a GPU box connected to your office network. Your team uses it from the browser. Only accounts, licences and the updates your administrator approves involve the cloud; prompts and documents are processed on the box.
Your browser
“What are the key risks in this contract?”
Supplier agreement.pdf
Local request
Inside LionBox
100% localCompany knowledge
Local AI model
Your answer
Local response
Answer ready, with sources
Setup for non-technical teams
From sealed box to first prompt in minutes.
Three guided steps. No specialist deployment and no software to install on employee laptops.
Inside every request
What happens after you press Send.
The browser, permissions, knowledge search, model and response path all stay within your office network and LionBox hardware.
No external inference APINo cloud prompt historyPermission-aware knowledgeLAN-only response path
Data path: laptop → office network → LionBox, and back the same way. Only accounts, licences and administrator-approved updates involve the cloud.
Local performance architecture
One GPU. Multiple private workstreams.
A dedicated high-capacity NVIDIA GPU runs our optimized build of Qwen 3.5. LionBox schedules concurrent requests across available compute, keeps user sessions separate and assembles the right context for each answer.
LionBox mobile app
One app to run the whole box.
The administrator app guides the technical work in plain language. Configure, monitor and update LionBox without turning everyday operations into an IT project.
Selected view
Network & model
0:001:05
The chat and the app, in one minute
Questions answered in the private browser workspace, then a look at the box from the LionBox phone app.
0:001:29
The unplug test
We disconnect the internet and keep asking. Every answer still comes from the box on the desk.
Local browser workspace
Your team’s daily AI workspace.
Chat, attach files, search approved knowledge and revisit source-backed answers in the browser - without installing desktop software. Share knowledge across the company or your department, and keep your own files private.
The AI model
One model, built for this box.
LionBox launches with our own optimized build of Qwen 3.5, tuned in-house for the LionBox hardware. More optimized and specialized models will follow as LionBox+ updates.
Available at launchQwen 3.5 · LionBox build
Optimized and tuned in-house for the LionBox hardware
- Inputs
- Text + images
- Runs on
- The LionBox GPU
- Status
- Preinstalled, ready to use
More models over time
- Optimized and specialized models, added through LionBox+
- Each one tested on LionBox hardware before release
- Installed by the administrator from the mobile app
Common questions
LionBox V1 is the on-premise private AI server behind this page. The security architecture documents every data path, and the box setup guide shows the installation step by step.
Where are prompts processed?
On the LionBox in your office. A request travels from the browser over the office network to the box, is answered by the model on its GPU and returns the same way. No external inference service is part of this path.
What happens when the internet is off?
Chat and indexed documents keep working on the office network. Optional internet search and downloading new updates need a connection; everything else runs on the box.
How do updates reach the box?
The administrator app downloads signed bundles and transfers them to LionBox over the local network. LionBox verifies each bundle before installing it; it does not fetch updates itself.
How many people can use it at once?
LionBox is designed for teams of up to 30 people. In our tests, replies started in under 0.5 s with 16 simultaneous sends.
Measured on LionBox V1
Speed you feel. Headroom you can trust.
More than 2,000 benchmark messages across realistic office work, simultaneous users, surge traffic and long documents.
0
failed requests
No errors, dropped accepted requests or memory failures across more than 2,000 benchmark messages.
0.07 s
Typical wait before the first word
The interface starts responding in under a tenth of a second during everyday use.
~2,000words/min
Delivered to each active user
Around eight times faster than average human reading speed.
<0.5 s
Wait with 16 simultaneous sends
Even when 16 people press Send together, every reply begins in under half a second.
5.2 s
To read a 90-page report
LionBox processed roughly 23,000 words and returned an answer in just over five seconds.
Your AI. Your advantage.
Claim your spot in the future of private AI.
Get your Founders Pass for €5 + VAT. Save €500 + VAT at launch. Free Shipping Worldwide.
Keep me in the loop
Be first to hear.
Get the launch announcement and delivery updates in your inbox.
Founders Edition · Free Shipping WorldwideA Founders Pass.
A Founders Pass.
A €500 head start.
€5 + VAT · unlock your €500 discount
Launch week is 7-13 December 2026, limited to 50 units. Your €5 + VAT will be deducted from the final price.
Assembled in Europe
LionBox V1 · Founders Edition
€5 + VAT Founders Pass · €500 launch discount