Distinguish the model, the system runtime and task-specific APIs.
1. Model and runtime
Gemini Nano is an on-device model. AICore supplies underlying system support on compatible devices. Your app integrates through supported APIs rather than assuming the model is a file bundled with every phone.
2. Task-specific GenAI
ML Kit exposes capabilities such as summarization, rewriting and image description. Select an API based on the task and its current support matrix; not all devices support every capability.
3. Local processing boundary
Inference can happen locally, but analytics, crash reports and application backups can still transmit data. A local-only promise must describe the entire feature, not just model execution.
Worked scenario
A diary app uses local summarization but logs the raw input in crash telemetry. The inference path is local while the overall privacy promise is broken.
Apply it
Map every place diary text can leave the process. Remove unnecessary transmission and persistence.
Check your understanding
You can explain the relationship between Nano, AICore and ML Kit without treating them as interchangeable names. Explain the decision and show evidence from your implementation or design. If you cannot demonstrate it yet, revisit the relevant section before continuing.