Direct answer
Codex model routing is the practice of selecting a model tier and account reference before a coding session begins. The routing decision should account for repo sensitivity, task type, expected token burn, regional policy, and whether the team is close to a model or account limit.
When this matters
- Low-risk UI copy edits should use a fast, economical model lane.
- Payment, auth, or migration work should route to a stronger model and stricter approval profile.
- A team wants to preserve premium model quota for production incidents and complex refactors.
Operating steps
- Classify repositories by risk and ownership.
- Define task classes such as read-only review, file write, deployment, secret access, and migration.
- Map each class to a Codex model lane, budget cap, and approval requirement.
- Expose the selected route in the local CLI before the session starts.
- Log route decisions so budget and incident reviews can reconstruct the choice.
Common risks
- Manual model selection can drift as teams add projects and agents.
- Overusing the strongest model can waste budget and hit limits early.
- Under-routing a high-risk task can create poor review quality or unsafe automation.
How AISwitchboard fits
AISwitchboard gives Codex teams a policy table for model routing and a local config package that applies the selected route without uploading raw secrets.