Model Routing
Different tasks suit different models. Model routing is the practice of choosing which model each role uses, and assigning it so the right reasoning is applied to the right job. Because assignment lives in the agent registry, you can route a role to a different model live, without a restart or redeploy.
Where the catalog lives, and where the routing lives
Models are configured in the platform's AI settings (the Coder dashboard), where you enable the models you want available. OpenFlows then references those models for each role - so the catalog lives in the platform, and the routing/assignment lives in your team's registry. To add a new model to the catalog, see Extending: Models.
Assigning a model to a role
- Make sure the model is enabled in the platform's AI settings.
- Open the agent registry for the role you want to change (for example, the builder).
- Set the model on that role.
- Save. The change is validated and takes effect on the team's next cycle. No restart, no redeploy.
Which model for which job?
A sensible starting point for routing models across the team:
| Role | A sensible model |
|---|---|
| Builder (FORGE) | A strong general coding model; does the bulk of implementation. |
| Reviewer (SENTINEL) | A model you trust for rigorous, adversarial reasoning about security and quality - often worth a higher-cost, more careful model. |
| Orchestrator / devops / writer | A reliable model; these roles are lighter on raw reasoning. |
You're free to tune from there based on what you observe.
What you should see
- The role now uses the assigned model for its work.
- The change applied without a restart; the next workspace provisioned for that role reflects the new model.
- The role's model shows in the control panel.
Availability notes