I kept running into the same tradeoff building cost tooling for teams shipping LLM features. Every attribution tool in this space works the same way: you point baseurl at a proxy, and it sees every call before it happens. That's genuinely useful if y...