Private Deployment AI Platform: Your Server or Ours
Two Shapes, One Boundary
Private deployment means either full on-premises operation — where your team installs, updates and monitors the platform on your own servers — or hosted operation, where WingPhoenix manages infrastructure in certified EU-based facilities. In both cases, the AI model weights, inference engine and configuration files are delivered as container images or bare-metal binaries. The boundary is clear: WingPhoenix owns the hardware stack and firmware; you own the runtime environment, access control and data lifecycle policy. This distinction governs everything from AI Robots fleet management to E1 learning robot classroom orchestration.
What Each Party Operates
Under private deployment, WingPhoenix delivers validated firmware, Docker Compose manifests and API documentation. You operate the Kubernetes cluster or VM pool, manage TLS certificates, apply security patches and define RBAC roles. Under hosted deployment, WingPhoenix operates the physical servers, network segmentation and patch cadence — but you retain full control over model versioning, prompt templates and log forwarding destinations. Neither model includes managed database services or SSO integration; those remain your responsibility, as outlined in our OEM/ODM services scope.
Data Residency Is Not Automatic — It’s Configured
Data residency depends on two choices: where the inference endpoint resolves, and whether telemetry is enabled. With private deployment, no data leaves your network unless you configure outbound HTTP calls to external analytics or translation APIs. With hosted deployment, all inference requests route through our EU endpoints — but logs, metrics and error traces only persist locally if you disable remote telemetry at installation. That setting is enforced at container launch time and auditable in the factory test report. For H3 health smartwatch integrations, wellness guidance output is generated locally on-device; raw sensor streams never transit to the platform unless explicitly enabled.
Compliance Starts at the Hardware Revision
Certification applies to the hardware revision, not the deployment model. CE RED, FCC Part 15 and UKCA declarations are tied to the physical device’s radio module, power supply and enclosure — not how the AI layer is hosted. Every unit shipped carries EN 62368-1 safety certification, RoHS 3 and REACH SVHC screening. For K1 companion robot, laser classification (class 1 LiDAR) and EMC scope are fixed at production. You cannot retroactively extend compliance to new software features — which is why our how to vet a supplier guide stresses reviewing factory test reports before signing.
The Hard Limit: What This Platform Does Not Do
This platform does not perform real-time data masking, enforce cross-border transfer restrictions at the API layer, or generate automated GDPR Article 32 reports. It does not integrate with SIEM tools out of the box, nor does it support FIPS 140-2 cryptographic modules. Claims of “GDPR-ready” or “HIPAA-aligned” are unsupported — because no audit report covers those domains. If a supplier cites such alignment without naming the certifying body, the exact standard clause and the date of the last assessment, pause the procurement process. Review the factory test report and request the AI Learning & Robotics line’s risk assessment document before approving any build.
| Decision Point | Private Deployment | Hosted Deployment |
|---|---|---|
| Data egress control | Full — all traffic governed by your firewall and proxy | Limited — outbound telemetry must be disabled at install |
| Certification scope | Hardware only — no software or cloud compliance | Hardware only — same as private |
| Patch responsibility | Your DevOps team | WingPhoenix infrastructure team |
| Model update method | Container image pull or offline bundle import | Automated rollout via internal registry |
| Audit trail source | Your logging pipeline (e.g., ELK, Splunk) | WingPhoenix-managed logs — exportable on request |
| Support escalation path | Direct to WingPhoenix engineering via SLA | Same, but with added infrastructure triage step |
| Minimum order volume | MOQ applies per hardware SKU — see OEM/ODM services | Same MOQ — no reduction for hosted delivery |
- Confirm your network’s TLS 1.3 readiness before deploying private instances
- Require signed DPA and data processing inventory before enabling hosted telemetry
- Validate firmware signature verification against public keys published in factory documentation
- Test failover behaviour with your load balancer — not ours
- Review the AI wearables line’s battery UN38.3 summary before integrating power-sensitive edge nodes
Private deployment gives you architectural sovereignty; hosted deployment gives you infrastructure relief — but neither relieves you of data governance duty. To proceed, download the latest product catalogue, identify your target SKU, then contact our OEM/ODM team to book hardware revision testing and confirm firmware delivery timelines.
Questions buyers ask
Can I keep all user data on-premises while still using your AI platform?
Yes — with private deployment, no inference data leaves your network unless explicitly routed via your API gateway or logging pipeline. Training data never leaves your environment.
Does your hosted option meet EU data residency requirements?
Our hosted service runs in ISO 27001-certified German and Dutch data centres; however, the hosted model is not GDPR-compliant by default — you must sign a DPA and configure retention policies per your jurisdiction’s rules.
What happens if our sample fails EMC testing during pre-market validation?
We rework the hardware revision at no cost and rebook the full test scope (EN 300 328, EN 301 489, EN 62368-1) with an accredited lab — documented in the <a href="https://wingphoenixai.com/factory/">factory test report</a>.
What does “private deployment” not cover — and where should we push back on supplier claims?
It does not include zero-trust identity federation, SOC2 Type II attestation, or real-time data lineage tracking. No third-party audit report exists for those controls — and no supplier should claim otherwise without verifiable evidence.
Related: more on ai memory platform · buyer guides · customisation tiers