Route chat through a configured OpenAI-compatible upstream with a virtual key for each app. Trace reported usage by key and compare rate-table estimates when usage and a model rate are available.