Can AI agent assist run on-premise or in a private cloud?
Yes, if the vendor runs its models inside your environment rather than calling a public AI service. Contact center AI can run as a vendor cloud service, as a private model in your VPC, or on-premise. UWXP, the Upstream Works AI layer, runs in the cloud, with on-premise deployment on request, and never trains on customer data.
Where the AI can run
| Where the models run | Where conversations go | Training on your data | Fits |
|---|---|---|---|
| A vendor’s public AI service | To the AI provider’s cloud for every request | Depends on the provider; confirm opt-outs in the contract before deployment | Teams with no residency or sovereignty rule |
| A private model in your VPC | Stay in your cloud tenancy | No, when the model is private to you | Regulated teams already on a public cloud |
| On-premise models | Stay in your own data center | No | Sovereign, defense and government environments, and regions where cloud is not accepted |
| UWXP | Cloud, with on-premise deployment on request; natively inside Totalix or beside the platform you already run | No training on customer data | Teams that want AI in the cloud, or on-premise on request, on specialized small language models tuned to CX |
The model-training point follows the site FAQ on AI and customer data; UWXP facts from the UWXP product page, 22 October 2026.
When it fits
Your conversations carry regulated, citizen or patient data, your policy forbids sending it to a shared AI service, or you need every AI decision explainable and auditable.
When it doesn’t
You have no residency or sovereignty requirement and are content with your platform vendor’s built-in AI service.
How Totalix does it
UWXP is the Upstream Works AI experience layer. It runs natively inside Totalix or stands alone beside the contact center platform you already run, deployed in the cloud, or on-premise on request. It uses specialized small language models and private LLMs matched to each task, with policy-based governance, compliance monitoring and end-to-end observability, so every decision is explainable, traceable and reviewable, and it never trains on your customer data.
UWXP → · Totalix Complete → · On-premise contact center software → · Reducing failure demand → · Keeping AI agents under control → · Request a demo →
Questions buyers ask
Does on-premise AI mean a weaker model?
Not for contact center work. UWXP uses specialized small language models tuned to CX, matched to each task, rather than one general model.
Can UWXP run beside our existing platform?
Yes. It runs natively inside Totalix or stands alone beside the platform you already run, on the Coexist, Displace, Replace path.
Is customer data used to train the models?
No. UWXP does not train on customer data in any deployment.
Can the agent desktop run on-premise as well?
Totalix On-Premise arrives in January 2027. Until then, Dedicated Hosted gives a single-tenant instance placed to meet data-residency commitments, from 1 December 2026.
Can we bring our own AI model, such as Google Gemini, OpenAI or Claude?
Yes. In Totalix Complete, the Omni AI Hub lets you bring your own AI model, including Google Gemini, OpenAI, Claude, Microsoft, IBM and Amazon engines, alongside UWXP, and reports on all of them in one place. A public AI service still sends conversations to its provider’s cloud; UWXP can run on-premise, on request.