Hybrid Architectures

The mature edge architecture isn't 'on-device OR cloud' — it's a routing decision per request:

• On-device handles — the common case, the private case, the latency-critical case: wake words, autocomplete, photo search over personal data, short summaries
• Cloud handles — the heavy case, the rare case, the case needing more capability than fits: long documents, hard reasoning, anything needing frontier parameters
• A router decides — complexity classifier, confidence threshold, or explicit user choice picks the path per request

Apple's architecture is the reference: on-device models by default, Private Cloud Compute for what exceeds them — the boundary is a designed policy, not a failure mode.