BM in a Box

Ready-to-run appliance

A pre-configured Blue Mesh appliance on NVIDIA DGX hardware, built to run fully air-gapped. Agents, voice, retrieval, and model training, all of it inside your own walls, with no outbound connection required for the platform to work.

HARDWARE
NVIDIA DGX, specified with you
PRE-BUILT
Platform installed and configured before it ships
INSTALL
Racked inside your boundary
RUNNING
Agents, voice, RAG and training, disconnected
What arrives

What is in the box

The hardware

NVIDIA DGX, specified with you against the workloads you actually plan to run, and racked inside your boundary.

  • NVIDIA DGX
  • Sized against your workloads

The platform, installed

BM Studio, BM Architect and BM Oasis, with Enterprise RAG and Voice agents, configured before the box ships, so what arrives is already working rather than a set of installers.

  • The full platform, configured
  • Working on arrival

Agents, voice, retrieval and training

Everything the platform runs elsewhere runs here: agents on the canvas, voice agents on your telephony, retrieval over your indexed documents, fine-tuning on your own data. Voice agents connect to your telephony over SIP on your internal network; the appliance needs no path to the internet, only a path to your PBX.

  • The same platform, disconnected
  • No outbound connection required

The audit trail

Every decision and result recorded locally, under your retention rules. It is yours and it does not leave the box.

  • Logged inside the box
  • Your retention rules
From order to running

Specified, built, installed, then updated on your schedule

01 // Specify

One session on what cannot leave the building and which workloads will run. The hardware is sized against that, not against a price list.

02 // Build

The platform is installed and configured on the appliance before it ships, and checked as a running system.

03 // Install

Racked inside your boundary and brought up on your network, your team with ours, with no path to the internet required at any step.

04 // Update

New models and platform versions arrive as a signed bundle on media your team verifies, then applies as a scheduled change it reviews before it runs. Nothing changes underneath a live workload.

The trade-off

Air-gapped means air-gapped

An appliance with no path to the internet does not get quiet improvements. That is the point of it rather than a limitation we are working around, and it is the whole decision.

No automatic updates

New models and platform versions arrive as a deliberate, scheduled, reviewable update. Nothing changes underneath a running workload because a vendor shipped something.

No telemetry home

We cannot see your usage, your prompts, or your failures, which means diagnosis is something we do with your team rather than from a dashboard on our side.

Everything is logged locally

Every decision and result is recorded for compliance audit, in your environment, under your retention rules. The audit trail is yours and it does not leave.

One box is one unit

Where you need continuity through a hardware failure, that is a second appliance and a failover plan, specified with you.

Who it is for

Organizations where the data cannot leave: defence, regulated finance, healthcare records, and national infrastructure. If your data can sit in a private cloud, that is cheaper and easier, and we will tell you so.

Where to go next
  • BM Oasis The training workload this hardware exists to carry: fine-tuning on your own data, with the weights staying yours.
  • Governance Search across decades of legislative and policy archives, the public estate an appliance most often gets specified for.
  • Trust and security The air-gapped mode written for a security reviewer: what runs inside the box, what never leaves it, and how updates arrive.
// specify the box

Tell us what cannot leave the building

In one session we will work out whether you need an appliance or a private cloud deployment, and specify the hardware against the workloads you actually plan to run.