This host
Run jobs on the machine the platform is installed on — ready as soon as it is installed.
VersaDesk Lite | Private AI computing platform
From one workstation to a whole GPU cluster
One platform, one way of working. Start using AI on a single machine; when you need more computing power, add other hosts or build a Kubernetes cluster. Computing and data stay inside your company.
- One workstation
- Many hosts
- Kubernetes cluster
Illustration
One way of working
Three kinds of computing, used the same way
Administrators connect the machines; users simply create a project, press Run and read the results. Each job runs wherever its project points.
Other Docker hosts
Turn spare servers into job runners from the maintenance console, and the platform sends work to them.
Kubernetes cluster
Build one or connect the cluster you already run; every project gets its own space and quota.
Users see exactly the same screens: the same form, the same results, the same terminal.
Build a cluster
A cluster in a few clicks from the console
No config files, no commands to memorise. Pick the machines, assign roles and press Create — the platform installs Kubernetes, the GPU components and shared storage in one go.
-
Pick machinesPick machines -
Master Nodes and VIPMaster Nodes and VIP -
Name and storageName and storage -
Final checkFinal check -
CreateCreate -
DoneDone
Pick machines
- Offline installation
- Rolling upgrades
- Backup and restore
- Add or remove nodes any time
Connect a cluster
Already have a cluster? Connect it
Paste an administrator's settings once. The platform sets up its own working area in the cluster and stays inside it, leaving your existing services alone.
-
A space per project
Each project gets its own space in the cluster; projects can't see each other.
-
Quotas set by administrators
Administrators decide how much CPU, memory and GPU each project may use.
-
Results come back on their own
Job outputs land in the project folder, just like on a single host.
Kubernetes cluster
Your existing services
The platform's area
Project
Project
Project
GPU
Share a single GPU
Administrators decide how many cards each project may use and whether it has them to itself or shares them; cards that support MIG can be split into several parts. One screen shows usage on Docker hosts and clusters alike.
Exclusive
A whole card for one project.
Shared
Projects take turns; the platform queues the jobs.
MIG partitions
One card split into parts for several jobs at once.
GPU Monitor: graphics cards on Docker hosts and clusters on one page
Host apps
Many hosts, one place to manage them
No more logging in to each machine. Check the status, containers, files and terminals of the platform host and every other host from the browser.
Host Status
Host Status
Live CPU, memory, disk and GPU.
Host Docker Manager
Host Docker Manager
Browse and clean up images and containers.
Host Terminal
Host Terminal
Open a shell as an account on that machine.
Host File Manager
Host File Manager
Upload, download and organise files.
Shared storage
Your data follows the job
Register a NAS once in the console and every host and cluster node mounts it. Whichever machine runs the job, it reads the same project folder.
What you already know
And everything else, as easy as before
Install AI packages from the App Store in one click, choose the desktop or classic view, and train models in the browser.
AI TRAINING
Real AI training
Every screen below comes from packages actually running.
Fine-tune a language model with LLaMA-Factory, follow the run as it trains, and test it in chat when it is done.
01
Choose a model and a dataset, press Start, and the training curve updates live.
WHERE IT RUNS
Runs on workstations, servers and DGX Spark
VersaDesk Lite installs on your own Linux machine, x64 or arm64. With an NVIDIA GPU it runs AI training and inference, and your team connects from the browser.
Workstation
x64 · NVIDIA RTX GPU
One workstation with a GPU lets a small team share AI tools.
Server
x64 · multiple GPUs
On a multi-GPU server the platform assigns GPUs to each project.
DGX Spark
arm64 · GB10 unified memory
A desktop AI machine; the platform recognises it and applies its own settings.
See VersaDesk on DGX SparkUbuntu and Debian (x64 / arm64) · Docker set up by the installer · General tasks run without a GPU
EASY SETUP
One installer, a few steps to go live
Run one command on the machine, then finish the setup wizard in the browser. The database and other services are created by the platform; there is nothing else to install.
admin@ai-server:~$
sudo sh versadesk-lite-linux-x64.run
- ✓ Check the machine architecture
- ✓ Prepare Docker
- ✓ Install the platform and maintenance console
- ✓ Start automatically at boot
→ http://192.168.1.10:8080
Open this address in a browser
Run one command on the machine; the installer does the rest.
-
01
Download the installer
One file each for x64 and arm64, holding the platform and the maintenance console.
-
02
Run one command
The installer checks the environment and completes the setup.
-
03
Finish in the browser
The wizard creates the administrator and sets up the site and network.
-
04
Update online
New releases install from the web interface.
YOUR WORKFLOW
Beyond the official packages: define your own
Official training packages or workflows not quite right? Write your own Runbook with your own image, script and parameters; the platform builds the run form from it and runs it on the same GPUs.
01
Define inputs and steps in the task editor.
Grow with you
Start small, grow any time
No need to buy everything up front. Start with the machine you have and add more when you need it — on the same platform all the way.
-
Single host
One workstation or server — for small teams and trials.
-
Several hosts
Add spare servers and spread the work.
-
Your own cluster
Build a Kubernetes cluster from the console — for departments sharing resources.
-
Connected cluster
Connect the cluster IT already runs; the platform stays in its own area.
Start with one machine, grow to a whole cluster
Ready as soon as it is installed; add hosts or a cluster when you need more.
Talk to us