01CSTM-D
Pilot on real data
Prove search, extraction or an agent on your own documents before you commit to production hardware.
CSTM-D · REV AcstmAI™ · Hardware
An AI workstation for one team or one pilot.
cstmAI™ Desk is a desk-side system with one or two professional GPUs, sized for a proof of concept, a department, or a lab that wants models running next to the work. It arrives burned in, with the cstmAI Platform installed and your first use case loaded.
Fig. 01 · Instrument assembly at the bench, NASA Kennedy
Most custom AI projects should start small and close to the people who will judge them. A workstation-class machine runs a capable open-weight model for a team of a dozen or two, needs no server room, and at the lower end of the range plugs into an ordinary office circuit.
We choose the chassis, GPUs, memory and storage for the model you plan to run and the number of people who will use it. If the pilot earns a production budget, the same software moves to a cstmAI™ Rack without rework: models, prompts, evaluation sets and connectors carry over.
| Form factor | Desk-side tower; 4U rackmount on request |
|---|---|
| GPUs | 1–2 professional workstation GPUs |
| GPU memory | 24–96 GB per GPU |
| System memory | 128–512 GB |
| Storage | 2–16 TB NVMe |
| Typical models | 7B–70B parameters (quantized at the upper end) |
| Concurrent users | About 1–25, depending on workload |
| Power | Roughly 0.8–2 kW; a dedicated circuit at the top of the range |
| Setting | Office, lab or small server closet |
Typical ranges. The quote names every part.No list prices
Front elevation
The drawing before the build.
A representative elevation. The chassis we specify varies by vendor and generation; the callouts don't.
Fig. 1 CSTM-D · REV A · Front elevationNot to scale
01
GPU cards, 1–2 (hidden)
02
Front I/O and power
03
Intake grille
Where it fits
Small enough to sit in the room. Big enough to prove the point.
02CSTM-D
One department
A shared model and document search for a single team: legal, engineering, finance or operations.
03CSTM-D
Build and evaluation box
Where engineers (yours or ours) iterate on prompts, evaluation sets and fine-tunes without touching production.
04CSTM-D
Single-site sensitive work
A machine that stays in one room, for data that should not travel even inside your own network.
Reel 02 · Lathe aisle seen from the catwalk, Paterson, NJ1994
How it's delivered
Four steps, each one signed off.
01
Spec
A workload interview: which model, how many people, what data. We write the configuration sheet.
02
Build & burn-in
Assembled, stress-tested under sustained GPU load, imaged with the cstmAI software.
03
Install
Delivered, networked, connected to your single sign-on, first use case loaded.
04
Hand over
Runbook, admin walkthrough and a named contact for support coordination.
FAQ
Questions we hear first.
Can a workstation run a useful model?
For many workloads, yes. Two 48 GB-class GPUs hold a 70B-parameter model at 4-bit quantization, and 8B–32B models run comfortably on one. We test your actual task on candidate models during discovery and show you the scores before you choose hardware.
What happens when we outgrow it?
The software is the same on every cstmAI configuration, so moving to a Rack is a hardware change, not a rebuild. Many teams keep the Desk afterward as their development and evaluation machine.
Which GPUs go in it?
Whichever fit the workload and are available when we order. We spec across vendors and generations; for workstations that usually means current professional cards with 48 GB or more of memory. The quote explains the trade-offs.
Get a quote
Spec your system.
Tell us the models you want to run, how many people will use them and where the hardware should live. An engineer replies with a first configuration and the questions that decide the quote.