Skip to contentAI that installs on your own infrastructure. Nothing leaves your network.Request a demo ↗

For the IT department

What it takes to host Oleron.

Oleron runs on your own infrastructure, with no internet access required. We designed the installation to graft onto what you already run, with as little friction as possible. Here is the outline; exact sizing is settled with your teams.

Installation

Installs the way you work.

From the lightest option to the most turnkey. You choose the format that fits your estate and the way your teams work, not the other way round.

Guided CLI installer

One command on a Linux server: it lays down the stack, configures internal TLS and creates the first administrator account. The simplest way to start.

Containers

A docker-compose bundle, or a Helm deployment on k3s if your teams are tooled for Kubernetes and want high availability.

VM / OVA image

A prebuilt image to import into VMware, Proxmox or KVM. Everything is preloaded, models included.

Sealed appliance

A preconfigured server, air-gap ready, delivered turnkey. Installation handled for your IT department.

Hardware

Two tiers, depending on your use.

A server with a professional GPU covers most cases. We size it precisely with you, based on the number of rooms and the level of summarisation expected.

Pilot tier

Proof of value, one or two rooms

GPU
Professional 24 GB GPU with ECC memory (e.g. NVIDIA L4 or RTX A5000)
CPU
8 cores min, 16 recommended
RAM
32 GB min, 64 GB recommended
Storage
512 GB to 1 TB NVMe, encrypted
Network
Internal LAN, no internet

Max quality tier

Production, multi-room, top-end summarisation

GPU
Professional 48 GB ECC GPU (e.g. NVIDIA L40S or RTX A6000). 80 GB for simultaneous multi-room use
CPU
16 cores min, 32 recommended
RAM
64 GB min, 128 GB recommended
Storage
2 to 4 TB NVMe (RAID advised), encrypted
Network
Redundant LAN possible, no internet

We use professional-class GPUs (ECC memory, drivers certified for continuous operation). Consumer cards are ruled out: their licence forbids use in a server room, and they do not offer the reliability guarantees expected in healthcare. Exact sizing and network details are settled during a scoping session with your CISO; they are not fixed on this page.

Software base

Nothing exotic to deploy.

A standard Linux and a GPU driver. Everything else lives inside the solution, isolated and reproducible.

Linux LTS: Ubuntu 22.04 or 24.04, RHEL / Rocky 8 or 9, or Debian 11 / 12
Recent NVIDIA driver (550+ branch) and a container runtime (Docker or containerd)
The rest (CUDA, models, services) ships inside the solution, nothing else to install on the host
Active Directory / LDAP integration, or OIDC / SAML SSO if you have an internal identity provider

Network

Internal, and that is all.

No outbound traffic, no telemetry. A few ports open on your internal network are enough. A full air gap is possible.

PortUsage
443Internal web interface (HTTPS)
22Administration (SSH)
636Active Directory lookups (LDAPS)
123 / 53Internal NTP and DNS

Updates come as signed packages that your teams apply themselves, on physical media in air-gapped environments. You stay in control from end to end.

We size it with your teams.

A scoping session to match the hardware and the installation to your actual infrastructure.