I kept finishing coding sessions with a small idea what I'd just spent on API calls :'D. The dashboards from the providers lag, and none of them give you a per-call breakdown tied back to your own code. So I built Tirith; an open-source CLI that sits as a transparent proxy between your code and any AI API. One import to add, one to remove. Drops in with a single line; no config changes Works with any provider you're already using (OpenAI, Anthropic, whatever) Tracks cost per call in real time Everything stays local on your machine; no data leaves your laptop Wrappers break when providers ship new endpoints or change their SDK. A transparent proxy just forwards requests, logs what it sees, and gets out of the way. Provider-agnostic by design. GitHub: [https://github.com/joedaviesio/tirith] Launch video: [https://youtube.com/shorts/uZCAqzqLfi4?si=k311z4DUVO2bwqYa] MIT licensed. Feedback very welcome; especially edge cases where the proxy behaves weirdly. First release, so I'm expecting rough edges.