Renderboxes Molecule Air Workstation
Renderboxes Molecule Air Workstation

Private AI, on your own hardware

RBOS runs AI models on your premises: your data stays in the building, and your team uses AI without a per-user cloud subscription. Built on Ubuntu LTS or Rocky Linux 9.

EXPLORE AI WORKSTATIONS

Why run AI on your own hardware

Data sovereignty

Your data stays in the building.

Prompts, documents and outputs are processed on the machine, on your network and under your control. Nothing is sent to a third-party AI provider.

Private by default

Only your team has access.

Every user signs in with their own account and two-step sign-in, and an audit log records who did what, and when.

No per-user AI subscriptions

One machine for the whole team.

Models run locally, so there are no per-user monthly fees or per-request charges. One machine serves the whole team over the local network.

RBOS Studio

LOCAL AI FOR THE WHOLE TEAM

An optional install that turns the machine into a private AI system for your team: chat, search over your own documents, a coding assistant and transcription, managed from a browser on the local network.

Every GPU, assigned individually

Each GPU is listed by vendor, memory and live use, identified by its PCI bus address rather than an index, and assigned to applications one card at a time.

RBOS Studio allocation board showing AMD and NVIDIA pools with per-card assignment

Reconfigure in one action

Save an allocation as a scene, such as a standard working configuration or a rendering configuration, and switch the whole system between them in one action. Users in session are notified, with a countdown before any service restarts.

RBOS Studio scenes with an active-session restart notice

Local AI tools, set up for you

A team chat assistant, search over your own documents, a coding assistant and transcription, each set up by a runbook that guides you step by step, without the command line.

RBOS Studio playbooks store

Know what fits before it loads

Curated models are checked against the machine before loading and marked as a fit, a tight fit or unsupported, with live memory meters for each GPU pool.

RBOS Studio model directory with fit verdicts and VRAM meters

Accountable by design

  1. Accounts

    Every user has their own account, so activity is attributable to a person rather than a shared login.

  2. Roles

    Administrators deploy and allocate; operators run what they have been granted.

  3. Two-step sign-in

    Authenticator codes, with recovery codes.

  4. Audit log

    A record of who did what, and when, visible to administrators.

  5. Encrypted access

    RBOS Studio is served over HTTPS on the local network.

Your data stays on the machine

Prompts, documents and model outputs are processed on the machine. The only outbound connections are the Renderboxes signed update channel and model downloads an administrator starts.

How RBOS is built

RBOS

The operating system.

A hardened Linux base configured for Renderboxes hardware, with AMD and NVIDIA GPU support, provisioning, desktop configuration and signed updates from Renderboxes.

SysControl

Included as standard.

Whole-system health in a browser over HTTPS: fans and temperatures, GPU presence, PCIe link status against the expected baseline, RAID health, platform sensors, and liquid-loop and leak monitoring with a controlled shutdown if coolant overheats, and email alerts.

RBOS Studio

An optional install.

A browser-based control plane for local AI. It assigns GPUs to applications, saves working configurations, deploys tools with runbooks, and manages accounts and access.

Two GPU vendors in one machine

MIXED-VENDOR GPU

RBOS detects every AMD and NVIDIA card on the bus and keeps the two vendors in separate pools, because their memory is not interchangeable. Large-memory AMD cards keep language models resident; NVIDIA cards run the creative and research software written for CUDA. Each card is assigned to the workload it is best suited to.

AMD Radeon AI PRO R9700 graphics card
NVIDIA RTX PRO 6000 Blackwell graphics card
NVIDIA RTX PRO 6000 Blackwell and AMD Radeon AI PRO R9700 graphics cards

Ubuntu LTS or Rocky Linux 9

Both are configured for Renderboxes hardware and include SysControl.

RBOS Ubuntu LTS RBOS Rocky 9
For AI, rendering and general Linux work Organisations whose security baseline requires the Red Hat family
SysControl Included Included
Desktop configuration Included Included
RBOS Studio Optional Optional
Packages and security apt, AppArmor dnf, firewalld, SELinux

Specification

  1. Base

    Ubuntu LTS or Rocky Linux 9, hardened and configured for Renderboxes hardware.

  2. Updates

    A signed Renderboxes update channel.

  3. GPUs

    AMD and NVIDIA in the same machine, each addressed individually.

  4. Monitoring

    SysControl: fans, temperatures, GPU presence, PCIe links, RAID, platform sensors, liquid-loop and leak monitoring, with email alerts.

  5. Access

    A browser on the local network, with no client software to install.

  6. Applications (RBOS Studio)

    Containerised, with installation, updates and clean removal managed for you, and your own applications alongside.

Common questions

Choose RBOS Ubuntu LTS for AI, rendering and general Linux work. Choose RBOS Rocky 9 where your organisation’s security baseline requires the Red Hat family. Both include SysControl, and both support RBOS Studio.

Not for the models you run on the machine: local models have no per-user or per-request charges. Some teams keep a cloud subscription for specialist tasks, but RBOS Studio does not need one.

Each vendor brings different strengths. AMD’s large-memory cards provide the most cost-efficient GPU memory capacity, which suits language models. NVIDIA cards run the extensive range of creative and research software built on CUDA. Together, both pools remain productive, each on the work it is best suited to.

No. Their memory is not interchangeable, so RBOS keeps the pools separate and never adds them together. Both pools work at once, on different jobs for different users.

It is refused before it loads, and RBOS Studio explains why and what to change.

Prompts, documents and outputs are processed on the machine. The only outbound connections are the Renderboxes signed update channel and model downloads an administrator starts.

Yes. Containerised applications can be added alongside the supplied ones, and models can be brought in from outside the curated set.

On the machine itself. Physical access to the machine is the proof of ownership; there is no reset from outside it.

Private AI on infrastructure you own.

CONFIGURE A SYSTEM