01CSTM-SW
Model serving
Several models at once, each request routed to the model that suits the task.
CSTM-SW · PLTcstmAI™ · Software
The on-premises AI platform under everything we ship.
cstmAI™ Platform is the base layer of every cstmAI system. It serves open-weight models, controls who may use them, and shows what they are doing. Search, agents and fine-tuning all run on top of it.
Fig. 01 · Inspecting infrared detectors, NASA JPL
Running a model is the easy part. Running it for an organization means single sign-on, roles, quotas, logs, health checks, model versions and an upgrade path. The Platform packages those, so your IT team operates one system instead of a pile of separate tools held together by a contractor.
It runs entirely on your hardware and depends on no outside model service. Your prompts, documents and any weights we train stay on your disks.
| Serves | Open-weight language, embedding, reranking, speech and vision models |
|---|---|
| Interfaces | Web chat for staff; an HTTP API for applications |
| Identity | Single sign-on through your identity provider; role-based access |
| Monitoring | Usage, response time and GPU health, with alerts |
| Runs on | Every cstmAI hardware configuration, Desk to Cluster |
| Network | Operates with no outbound internet connection |
Where it sits
Its place in the stack.
Every module runs on the same platform and hardware, under the same governance.
L4 · Modules
L3 · Control
What it handles
The parts every AI deployment needs and nobody wants to build twice.
02CSTM-SW
Access control
Sign-in with your existing accounts; roles and per-team limits on models and data.
03CSTM-SW
Monitoring
Who uses what, how fast it answers, and whether the GPUs are healthy.
04CSTM-SW
Model lifecycle
Versioned models, staged rollouts and rollback when a new version scores worse.
Reel 02 · Milling and boring area, Paterson, NJ1994
How it's delivered
Four steps, each one signed off.
01
Install
Loaded and tested during hardware burn-in, before delivery.
02
Connect identity
Joined to your identity provider; roles mapped to your groups.
03
Load models
The models chosen in discovery, with their evaluation scores recorded.
04
Hand over
Runbooks and admin training for your IT team.
FAQ
Questions we hear first.
Which models can it serve?
Open-weight models such as Llama, Qwen, Mistral, DeepSeek, Gemma and gpt-oss, plus embedding and speech models. We pick and test models for your tasks. License terms differ by model, and we review them with you.
Can our IT team run it without you?
That is the aim. Handover includes runbooks and admin training. If you'd rather we keep it running, managed operations is available.
Does it need internet access?
No. It runs entirely on your hardware, and updates are applied on your schedule.
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.