Model Routing

Saga lets you assign a different model to each generation task, rather than one global default.

Tasks you can route independently

• epic_generation
• story_generation
• story_refine
• story_splitting
• subtask_generation
• agent_prompt
• agents_md

Configuring routing

Open Settings → Model Routing. Each task shows a live model picker populated from every enabled provider's listModels() call. Pick a specific model, or leave it as "auto" to let Saga pick a sensible default for that provider.

How it's stored

Routing lives in config.yaml as a plain model ID string or the literal "auto":

ai:
  routing:
    epic_generation: "auto"
    story_generation: "claude-sonnet-5"
    subtask_generation: "auto"

An older { tier: ... } object form is still read for backward compatibility and coerced into the string form automatically.

What shows during generation

Every generation run displays <model-id> (<provider>) in both the progress notification and the Generation Review panel header, so you always know exactly which model produced what you're looking at.