Dual Model

A fast Local model answers most things well, but the occasional hard question wants a stronger one. Dual Model lets you keep both without choosing up front: simple messages stay on your fast default model, and only the complex ones are routed to a stronger secondary model. Loca makes that call per message, so you never switch by hand.

How the choice is made

Before answering, Loca checks how involved your message is and routes it accordingly. That check is itself a prompt, the Complexity Check, which lives in the same Prompts group as your system prompt, so you can see and adjust the rule that decides what counts as complex. A short, factual question stays on the fast model; a multi-step or open-ended one goes to the stronger one.

The point is to spend the stronger model only where it earns its keep. Simple turns stay fast and cheap on your default model, while the hard turns get the capability they need, without you watching the conversation to decide which is which.

When to use it

Dual Model pays off when your two models differ enough to be worth choosing between: a quick Local model for everyday replies paired with a more capable one, Local or External, for the hard cases. If you only run a single model, there is nothing to route to, and you can leave this off.

How It Works

The complexity check is a small extra turn that runs before the real one. Loca sends your message to a model with the Complexity Check prompt, which asks for a single quick judgment of how involved the request is, and reads the answer as a score. That score decides which model handles the actual reply: below the line it stays on your default, above it the turn is handed to the secondary model. Because the judgment happens per message, a single session can move back and forth between the two models turn by turn, spending the heavier one only when a message warrants it.

With your models settled, the next chapters cover the other things that shape a reply, starting with talking to Loca by voice: Speech-to-Text.