Primary provider
Azure OpenAI
bp AI Gateway Control Plane - bp Sphere
Model independence, provider routing, fallback, and policy proof
ScaleGateway credibilityModel independence
AI Gateway Control Plane
This surface answers the workshop question that usually kills AI demos: is the runtime tied to one model vendor? bp Sphere uses governed routing, provider authorization, policy proof, and fallback posture so the operating model survives model swaps, outages, and security controls.
Fallback chain
Anthropic → Gemini → Rules
Policy proof
Required before execution
Blocked runs
3 unsafe prompts quarantined
Gateway routing proof
A skeptical bp leader should be able to see which provider was selected, why it was selected, what fallback posture is available, and whether execution was blocked or permitted under policy.
| Ingress provider | AWS |
|---|---|
| Selected provider | Azure OpenAI |
| Fallback provider | Anthropic |
| Emergency floor | Rules Engine |
| Policy resolution hash | pol-res-9e2b4a |
| Evaluation hash | eval-44cf11 |
What the operator can prove
This is the workshop version of model independence: not just saying multiple models exist, but showing provider authorization, governance coverage, and what happens when the preferred model path is unavailable.
| Approved providers | Azure OpenAI, Anthropic, Gemini |
|---|---|
| Provider authorization | AI gateway policy CR-17 |
| Blocked secrets / PII | 14 blocked this week |
| Fallback simulation | 2.4s provider cutover |
| Execution posture | Human-governed on fallback |
| Recovery state | Warm standby active |
Why this matters
bp does not need another AI demo that assumes one vendor is always available. This surface shows the runtime can switch providers without breaking governance, evidence, or the operator workflow.
Workshop proof points
Use the executive question console to ask how the runtime behaves if a preferred provider fails. Then show the operator still receives a governed answer with provider selection, fallback, and policy proof intact.