Make a deployment decision from requirements and measured constraints.
1. Cloud inference
Cloud services can offer capable models across many devices, but add network latency, data transfer and usage costs. Put privileged provider credentials behind a backend or a supported protected mobile integration.
2. On-device inference
Local inference can reduce data transfer and work offline, but depends on compatible hardware, model availability and memory. A model download can fail independently of the feature permission check.
3. Hybrid routing
A routing policy considers task support, connectivity and user data consent. Local failure is not permission to upload private data. Make the fallback explicit and expose a useful unavailable state.
Worked scenario
A private diary summary requires local-only processing. On an unsupported phone, manual editing is the fallback; silently switching to a cloud model changes the product promise.
Apply it
Create a decision table for diary summaries, public travel questions and receipt extraction. Include unsupported hardware and offline conditions.
Check your understanding
Your fallback respects the same privacy promise as the preferred route. Explain the decision and show evidence from your implementation or design. If you cannot demonstrate it yet, revisit the relevant section before continuing.