Model Usage

How Routing Works

How PrettiFlow automatically picks a model for each task.

PrettiFlow decides which model to use based on what kind of task it's doing. You don't configure this — it happens automatically on every request.

Task-based routing

Tasks fall into two broad groups:

Decision tasks

Understanding your prompt, building context, and updating the plan. These are quick and use fast, economical models.

Execution tasks

Generating and editing code, writing commit messages, and producing project details. These use stronger models for better output.

Automatic fallback

If the preferred provider for a task isn't available, PrettiFlow cascades to the next option automatically — so a single provider outage doesn't block your build.

Pick the best-fit model

Based on the task type, PrettiFlow selects a preferred model.

Check availability

If that provider is configured and reachable, it's used.

Fall back if needed

Otherwise PrettiFlow moves to the next provider in line, keeping your build running.

Safety checks

Before running potentially risky operations (like shell commands), a separate lightweight "judge" model reviews the action first. This guardrail runs independently of the model doing the main work.

The result: fast, cheap models on light work; strong models on heavy work; and automatic recovery when a provider is unavailable — all without you choosing anything.
Copyright © 2026