This is neat — curious how you’re detecting quota usage across different providers (OpenAI, Anthropic, etc.). Are you parsing responses heuristically, or is there a library?

Also wondering how accurate this stays as providers keep changing token accounting.