Self-Hosted AI Server Setup
Use this guide when the customer runs self-hosted AI components on their own server and the host must be prepared before deploying vLLM or TEI.
This page keeps the host-level preparation in a separate place from the product installation scenarios.
Baseline used for the original setup
The original setup was tested on the following baseline:
Hardware
- architecture:
x86_64 - CPU:
16cores, exampleAMD EPYC 7302 - RAM:
48 GiB - GPU:
2x NVIDIA RTX A6000
Software
- operating system:
Debian 12.6or compatible - Docker:
27.0.3or newer - Docker Compose plugin:
2.28.1or newer - KVM virtualization
Treat this as a tested reference, not as the only supported combination.
Host prerequisites
Before installing self-hosted AI services, make sure you have:
- root access on the server
- Docker Engine installed
- Docker Compose plugin available as
docker compose - NVIDIA GPU available when the selected model requires GPU acceleration
If Docker is not installed yet, use the official guides:
NVIDIA GPU and container runtime setup
For Docker containers to access the NVIDIA GPU, install the drivers, CUDA packages, and NVIDIA container runtime on the host.
These commands are taken from the original server setup guide and are intended for Debian 12.
Step 1: Install required packages
apt update && apt install -y gcc ca-certificates curl gnupg net-tools tclsh unzip python3-pip software-properties-common wget
Step 2: Install Linux headers
Step 3: Add CUDA keyring
cd /tmp
wget https://developer.download.nvidia.com/compute/cuda/repos/debian12/x86_64/cuda-keyring_1.1-1_all.deb
dpkg -i cuda-keyring_1.1-1_all.deb
apt update
Step 4: Install CUDA
Step 5: Add NVIDIA signing key
Step 6: Add NVIDIA container runtime repository
curl -s -L https://nvidia.github.io/nvidia-container-runtime/debian12/nvidia-container-runtime.list | tee /etc/apt/sources.list.d/nvidia-container-runtime.list
Step 7: Install NVIDIA container runtime
Step 8: Restart Docker
Verify the host before deploying AI services
Before starting vLLM or TEI, verify:
nvidia-smidetects the GPU- Docker is running
docker composeis available- the server has enough GPU memory for the selected model
Then continue with:
Security practices
When exposing self-hosted AI services, also consider:
- Use strong and unique credentials for all exposed services.
- Restrict open ports with firewall rules.
- Keep the host and installed packages updated.
- Use HTTPS for external access.
- Apply access control and user management.
- Audit the server regularly.
- Use encrypted communication where possible.