Xephyr has partnered with Dicker Data to bring Xavier, our analytics agent, to the Cisco AI Rally Kit. Organisations can now run trusted analytics on their own data, on hardware they own, without moving that data to the cloud.
We announced the partnership at the Dicker Data partner event, where Lawrence Puang, our Founder and CEO, presented Xavier in the Partner Use Case Showcase with a telco demonstration recorded on the kit.
"Xephyr was the first ISV partner to leverage the Dicker Data Cisco AI Rally Kit infrastructure to design and build AI agents that accelerate real-world business outcomes."
Vickie Madeleine, General Manager Cisco Group, Dicker Data
Why this matters: most enterprise data isn't going to move
For the past few years, the default assumption has been that enterprise AI runs in a public cloud. For a lot of organisations, that assumption doesn't hold.
Telcos, utilities, health and government organisations run critical infrastructure and hold sovereign data that cannot touch the public internet. Many others have data spread across systems that can't easily move, because of regulation, security, contracts or sheer size. And every new AI use case adds to a cloud token bill that grows with every question asked.
We think enterprise AI is heading towards hybrid by default. Some workloads will run in the cloud. Many more will run right next to the data, on infrastructure the organisation controls. The AI goes to the data, not the other way round.
Until now, the hard part has been the infrastructure. Standing up AI on-premise usually meant a long project before anyone could ask a single question. The partnership with Dicker Data changes that.
Proven fully air-gapped at Dicker Data's Tech X
At Dicker Data's Tech X event in Melbourne on 29 September 2026, we ran Xavier fully air-gapped on the Cisco AI Rally Kit. No connection to the public internet. No data leaving the system.
Questions went in as plain English, from "how many calls did this customer make this month?" to "show me the top 3 sites by jitter", and came back as a number, a table, a line graph or a bar graph.
- About 60 seconds per question, answered on the kit's own hardware.
- 90%+ accuracy on the kit's open-weight model, with more optimisation to come.
What runs on the kit
Dicker Data distributes the Cisco AI Rally Kit to its partner network. Xavier runs on the kit as a ready-to-use workload, so partners can show customers a business outcome from day one.
| Layer | What it does |
|---|---|
| Cisco AI Rally Kit | The on-premise AI infrastructure, supplied through Dicker Data |
| NVIDIA GPUs and Nemotron | Four NVIDIA RTX PRO 6000 Blackwell GPUs run NVIDIA Nemotron, an open-weight model, so no request goes to a public model |
| Red Hat OpenShift | Serves the model with vLLM and runs Xavier's services |
| Everpure | Storage on the kit, so the data Xavier queries stays on infrastructure the organisation owns |
| Xavier | Answers business questions in plain English, runs the analysis and shows its working and source |
Sovereign by design, including fully air-gapped
Xavier runs inside the customer's environment, under their governance rules. Customers choose where:
| Deployment | Best for |
|---|---|
| Private cloud | Organisations that want cloud flexibility inside their own tenancy |
| On-premise, on the Cisco AI Rally Kit | Data that needs to stay on infrastructure the organisation owns |
| Fully air-gapped | Critical infrastructure and the most sensitive environments. No internet connection and no data leaving the building |
Because the compute belongs to the customer, the cost of each question is predictable. Customers pay for compute, not a per-token bill that rises with usage.
Where it makes the biggest difference
- Telecommunications. Call records, voice quality and device data often sit in separate systems, much of it on-premise. Separately from the kit, Xavier is live in production at a major Australian telecommunications company, where questions that took an analyst 2 to 4 hours are now answered in under 3 minutes.
- Utilities and critical infrastructure. Operational data that has to stay off the public internet.
- Health and research. Patient and research data that privacy and ethics rules keep inside the organisation's walls.
- Financial services and government. Risk and security teams can see exactly where the data and the model live, which makes approval far simpler.
Thank you
Thank you to the Dicker Data team for hosting the event and for their work getting Xavier running on the kit in their lab. Thank you to Cisco, NVIDIA, Red Hat and Everpure, whose technology makes up the stack. And thank you to our team, who made it happen while still delivering for our clients every day.
Want to see Xavier running on the Cisco AI Rally Kit? Speak to Xephyr or your Dicker Data Cisco BDM. Book a conversation with our team
Related: Xephyr named a finalist in the 2026 Financial Review BOSS Most Innovative Companies · Xavier · Telecommunications