Skip to main content

Cost Optimization

Budget Policy​

Configure actions when spending approaches limits:

providers:
  routing:
    budget_policy:
      soft_limit_action: downgrade  # Switch to cheaper model
      hard_limit_action: reject     # Block requests

Soft Limit Actions​

  • downgrade — Route to a cheaper model/provider
  • notify — Continue but alert operators
  • queue — Hold requests until budget refreshes

Hard Limit Actions​

  • reject — Block all requests (fail-closed)
  • queue — Hold requests

Cost Attribution​

Track spending by dimension in the console Cost Attribution dashboard:

  • Team
  • User
  • Project
  • Provider
  • Model
  • Task type

CLI Commands​

Check provider status and estimated costs:

kt secrets status
kt gateway check --verbose