One row per planning model. Highest priority runs first and automatically falls back on errors or rate limits.

Limits here apply to both planning and build traffic. Set provider/model RPM/TPM caps and monitor live usage.