我如何意外地构建了一个 LLM 成本追踪工具

Last month, I received an API bill that shocked me with a total of $2,847, and I had no idea where the charges came from. While building a side project, I utilized several services including OpenAI, Anthropic, Stripe, Supabase, and SendGrid, each with its own dashboard and billing page. This led to the frustrating routine of checking five tabs every morning to monitor any spikes in costs. To address this, I wrote a simple script to intercept outgoing API calls and log the associated costs. Interestingly, I discovered that my onboarding flow was generating 14 LLM calls per new user, which highlighted the source of the expenses.