Skip to content
ALDBRN
Launch updates
A MANUFACTURING PLANTPRODUCTION LINECONTROL ROOMTHE MACHINEYOUR NETWORKSILOS

Hardware

The platform runs on a machine you own.

Inside your building, on your network, under your own access rules. Three ways in, from a desk to a rack, and a pilot that starts before you buy anything.

See the three tiersGet launch updates

Manufacturing plant · Perseus

For IT and OT teams

Local AI Models in Industrial Environments

ON PREMISESYOUR WORKSTATIONYOUR GPUSINSIDE YOUR NETWORK
What crosses
Model weights come in during setup.Alerts or reports can be sent out.
Who owns it
Your IT team owns deployment.OT owns the information that goes in.
What it runs
Qwen3-30B-A3B,DeepSeek-R1-Distill-32B,Mistral-Small-3.2-24B,Gemma-4-31B,Recursive's Flint.
Specs
VRAM: 48 GB up to one H200 (141 GB).Context window: up to 256K tokens.

Why this matters

Open weights give you predictable infrastructure cost. The model's behavior is more consistent and only changes when you update it.Sensitive records and operational data stay on premises.

Feasible setup in November 2026

Three ways in, from a desk to a rack.

ALDBRN-RUNTHE MACHINEYOUR SITEYOUR NETWORK, OVER VPN

Start without buying

ALDBRN-run pilot

A machine ALDBRN operates, so the first program can start before a purchase order.

Best for
Proving the case at one site
Who owns it
ALDBRN. You know where it is.
What it runs
The full platform for one program and a handful of operators
GPU memory
Sized with you

What it means for you

Your records sit on a machine we run, in a place you know, and the first program is generated and reviewed there. Everything else about the deployment is the same, so moving to your own machine later changes where it runs, not what it does.

  • Proving the case before anyone buys a machine
  • A site without a server room
  • Seeing a program on your own records before committing
UNDER A DESKTHE WORKSTATIONGPU CARDSSIZED WITH YOU

One site or department

Desk workstation

A workstation under a desk, in a room you control.

Best for
One site, one or two programs
Who owns it
You
What it runs
The full platform: models, scenarios, labs, AI assistant
GPU memory
Sized with you

What it means for you

A workstation under a desk runs the full platform for one site: the models, the scenarios and labs, the AI assistant. It is the first machine most sites choose, and a department can own its instance outright.

  • One site, one or two programs
  • A department that wants its own instance
  • A program you can lock in a room
IN YOUR RACKROOM TO GROWALL YOUR SITESYOUR NETWORK

Whole organisation

Rack server

A server in your rack, for every site and every program.

Best for
Several sites, every program, room for larger models
Who owns it
You
What it runs
The full platform, with headroom
GPU memory
Sized with you

What it means for you

A server in your rack serves every site and every program from one place, under your own access rules, with room for the models that come after today’s. Your own node, in your own building.

  • An organisation rolling out to every site
  • Work that must stay inside your own walls
  • Room for the models that come after today’s

No part numbers and no prices here. We specify the machine with you, and the figures come from the quote, not the website.

The three tiers, side by side.

The three tiers compared.
TierALDBRN-run pilotDesk workstationRack server
Best forProving the case at one siteOne site, one or two programsSeveral sites, every program, room for larger models
ClassA machine ALDBRN operatesA workstation under a deskA server in your rack
GPU memorySized with youSized with youSized with you
Who owns itALDBRN. You know where it is.YouYou
How you startImmediately, on our machineAfter the machine is inAfter the machine is in

What leaves, and what stays.

The honest list. We’d rather tell you what crosses the boundary than be caught being absolute about it.

What leaves

  • Model weights, downloaded once at install.
  • Email notifications, to addresses you configure.

What stays

  • Your documents and records.
  • The scenarios, labs and programs built from them.
  • Every question and answer, and every memo.

A pilot can start on an ALDBRN-run machine instead, so you can begin before buying hardware. You know where it is, and the rest is the same.

YOUR PLANTTHE MACHINEPARTS INRECORDS, READ ONLYASSEMBLY LINEFINISHED ROBOTSWHAT LEAVESA ROOM IN YOUR PLANTTHE MACHINEYOUR OPERATORSYOUR NETWORKWHAT LEAVESEMAILWEIGHTS, ONCE

It reads. It never writes back.

Documents and records
Drawings, procedures, handover notes and change records, read where they already sit.
Read only
No connection writes anything back. It cannot change a document, a record or a piece of equipment.
Never a way to run the plant
A learning exercise never grants permission on a real site. Real changes belong to your authorised people and procedures.

How a program gets to your site.

Four steps with ALDBRN on hand, then we keep it current.

  1. Specify the machine together

    Desk workstation or rack server, sized with you. Or start on an ALDBRN-run machine.

  2. Connect the records, read only

    Drawings, procedures and change records, where they already sit. The platform reads and never writes.

  3. Generate and review the first program

    Local models build the scenarios and labs from your records. Our experts review them on your machine before anyone trains.

  4. Train with us on hand

    Your operators work through the program in a browser on your network. We stay close while they learn how to work with it.

Then we keep it current.

  • Models updated as better open weights ship.
  • Monitoring with alerting on the machine.
  • Backups, and upgrades that keep your configuration.

Questions about the machine

Where does our data live?

On the machine, inside your network. Your documents and records, the index built from them, the scenarios and labs, every memo and every question stay on its disks. No document and no question is sent to a model vendor.

Do you train on our data?

No. The models read your records to generate scenarios and to guide a learner. Nothing you give the platform is used to train a model, ours or anyone else’s.

So nothing ever leaves our network?

Two things do. Model weights are downloaded once when the machine is set up. Email notifications go to addresses you configure. Everything else stays in.

What does it read?

Documents and records: drawings, procedures, handover notes and records of what changed. Read only, where they already sit.

Can it change anything in our systems?

No. It reads and never writes. It cannot change a document, a record or a piece of equipment, and a learning exercise never grants permission on a real site.

What hardware do we need?

One of three: an ALDBRN-run machine to start without buying, a desk workstation for one site, or a rack server for a whole organisation. We size it with you, and the figures come from the quote.

What does getting started look like?

Specify the machine together, connect the records read only, generate and review the first program on your machine, and train with ALDBRN on hand. Pilots are open at selected sites.

Tell us where your records live.

Pilots are running at selected sites. Signing up asks to be considered. It does not reserve a place or a price.

Get launch updates