On-premise AI hardware
AI server in your server room
Sending company documents to an external API can be unacceptable legally or simply out of caution. We deploy an on-premises machine with 128 GB GPU memory running open-source language models with no cloud connection.
In practice
Data stays in the building
An on-premise LLM server is a single unit in a rack cabinet. It delivers the same performance as a cloud service, but without sending any data outside.
A configuration with 128 GB of VRAM runs quantized 70B class models today. That is usually enough for a dozen concurrent conversations in a mid-sized company.
We start with a single department, most often sales proposals or technical documentation. We measure the actual time saved before connecting additional teams. After a two-day training, your IT specialist adds models and creates accounts independently.
The hardware belongs to you from day one. If you change software providers after a year, the machine stays in place.
How it works
The machine arrives at your server room ready to work: operating system, drivers, and several open-source models already installed. Prompts and completions are stored on your local drive, and the default configuration blocks all outbound traffic.
Integrations
The server exposes a standard API, so we connect it to the tools you already use: email, CRM, network drive, spreadsheets. We mirror document permissions from your systems so no user sees files in responses they lack access to.
Who decides
The admin password stays with you, and your company remains the data controller. We only access the server when someone on your team grants us entry. The audit log tracks every query, and the SLA defines who answers the phone if hardware fails.
Problem
Concerned about sending company documents to the cloud?
HEXART solution
A server with 128 GB of VRAM sits on your premises and runs quantized 70B class models. Queries remain on your local drive, and access is managed by your administrator. We start with a single process and measure time saved before expanding the deployment.
Ask about server configurationQuestions and answers
Frequently asked questions
What models fit on a server with 128 GB of VRAM?
Quantized 70B class models, which is usually enough for a dozen concurrent conversations in a mid-sized company.
Do queries leave the company?
No. Prompts and completions are stored on the local drive, and the default configuration blocks all outbound traffic. The admin password stays with you, and your company remains the data controller.
Do we own the hardware?
Yes, from day one. If you change software providers after a year, the machine stays in place. The SLA defines who answers the phone if hardware fails.
Who manages the server after deployment?
Your IT specialist. After a two-day training, they add models and create accounts independently. The server exposes a standard API, so we connect it to the email, CRM, network drive, and spreadsheets you already use.
No-obligation call
Schedule a workshop with Paul
Thirty minutes or a full workshop, choose what suits you. Book a slot directly in the calendar and get instant confirmation.
Paul Lazniak
HEXART Founder
- Meeting always with the same person
- 30 min · 90 min · 15 min
- biuro@hexart.pl
- +48 662 016 430
Contact
Let us talk about your project
We will show you a working product before we start talking about it. Let us know what you need: a film, an XR environment, an AI system, or brand identity.
Write or call
A few sentences are enough: what needs to be created, by when and for whom.
- biuro@hexart.pl
- Phone
- +48 662 016 430
- Address
- Aleja Zwycięstwa 96/98, 81-451 Gdynia