A practical way to choose what runs a conversation.
Pick providers for ownership and routing. Pick models for the kind of thinking or agent work in front of you.
Think in two layers
Providers answer where access comes from. Models answer what kind of behavior you want.
Separating those decisions makes Atoi easier to manage, especially when you are comparing several families or rotating through experiments.
A simple selection rule
- Use smaller or faster models for quick iterations and light drafting.
- Use stronger reasoning models when the thread needs judgment, synthesis, or careful code review.
- Keep multimodal-friendly models available when your workflow includes images or mixed inputs.
- Prefer one default per task shape instead of changing models for every message.
What changes over time
Provider catalogs evolve. New model families appear, older ones are renamed, and performance tradeoffs shift.
Atoi is designed to handle that reality by keeping selection explicit and by letting you prune the list to something manageable.