On-Device Model Router (Apple FM)

Spend less and stay fast by sending each task to the smallest model that still works (on-device vs Private Cloud Compute vs custom Core AI).

Model router with cost/latency/quality rules using Dynamic Profiles and model modifiers.

Write rules that pick SystemLanguageModel, PrivateCloudComputeLanguageModel, or CoreAILanguageModel, then apply them with Dynamic Profile modifiers. iOS 27 and macOS 27 only.

Lessons

Key concepts

Route tasks to SystemLanguageModel, PrivateCloudComputeLanguageModel, or CoreAILanguageModel.
Rules live in app code. Dynamic Profiles apply .model and .reasoningLevel.
Pick the smallest Apple path that still does the job.
One active Profile at a time on iOS 27 and macOS 27.

Takeaways

You can pick the smallest Apple model that still works and fall back without breaking privacy.
You can write the rule in code and log the route key every request.
You can switch the model with a Profile modifier instead of rebuilding the chat UI.

Agent briefs

FAQ

Which Apple models can the router pick?

SystemLanguageModel on device, PrivateCloudComputeLanguageModel, or CoreAILanguageModel for a custom Core AI model.

Where do routing rules live?

In app code. Log the route key on every request so you can see which path ran.

How is the route applied?

Dynamic Profiles apply the .model and .reasoningLevel modifiers, so you switch models without rebuilding the chat UI.

What's the rule of thumb?

Pick the smallest Apple path that still does the job, and fall back without breaking privacy.

Which OS versions does the router need?

iOS 27 and macOS 27 only, with one active Profile at a time.