Why the cloud question stops projects
Geometry, formulations, and test results are the company's crown jewels, and the review that asks where they will be processed is where the pilot usually stalls. The question is rarely about the model itself. It is about the path the data takes, who else can reach it along the way, and what the contract says when something goes wrong.
Answering that with a deployment diagram rather than a policy promise is usually what unblocks the project.
What local execution means concretely
Training and prediction run on your own hardware, so in local setups the data stays on customer infrastructure rather than being uploaded for processing. Your simulation results, measurements, and geometry files sit where they already sit.
The model is built against them in place, and the predictions come back from the same machine. No copy of the training set has to be staged somewhere else for the workflow to run. See how STOCHOS works for what happens inside that step.
Air-gapped operation
STOCHOS supports fully air-gapped, on-premise operation using an offline license. In that configuration it makes no outbound network connections, so the entire workflow runs sealed. The default networked install performs an online license check.
Which of the two you run is a deployment decision, taken once, at install time. The readout below sets the two modes side by side.
What this changes about cost
With the agents running locally on your own GPU there is no token limit and no token cost, so the price of a workflow does not scale with how much you use it. That removes the usual reason to ration a tool.
An engineer can rerun a study, sweep a parameter again, or let a workflow explore a wider range without anyone counting calls. The hardware you already bought is the budget.
If your governance rules out external processing entirely, the offline license is the configuration to ask about. If they do not, the networked install is simpler to run and the data still stays on your hardware. Either way you can build the same workflows in STOCHOS Flow.
Common questions
01Is STOCHOS air-gapped by default?
No. The default networked install checks the license online. Fully air-gapped operation is supported through an offline license.
02Can I still use cloud services if I want to?
Yes. The stack is local first, with optional cloud services where you want them. In air-gapped mode there is no network, so cloud models are not available and local ones run instead.
03Who is on-premise AI for?
Defence, classified, and regulated-IP work, and any team whose data governance rules out external processing.
04What is on-premise AI?
On-premise AI runs on your own infrastructure instead of a vendor's cloud. Models train and predict on your hardware, and your data never leaves your network. STOCHOS and STOCHOS Flow are built local-first, and the assistant can use a local language model through Ollama.
05Does the AI assistant still work with the network sealed?
Yes. In the air-gapped configuration the assistant's local language model runs through Ollama and any knowledge bases stay indexed on your own machine, so it keeps helping you build and debug workflows. Cloud language models are the one part that needs the network, and those stay off.
STOCHOS
The predictive engine: fast predictions with quantified uncertainty.
STOCHOS Flow
Build the workflow as a node graph, run it where your data already is.
How STOCHOS works
DIM-GP in plain language, and why it suits small, expensive datasets.
News and Guides
Project write-ups, releases, and the rest of the guide series.
