Token Audit breaks every LLM API request into question / thinking / reply tokens and shows where your money goes. Reasoning models like o1 burn up to 77% of tokens on hidden thinking — at premium prices. The tool estimates costs across 14 models, fires alerts on wasteful patterns (high reasoning ratio, runaway automation), and its AI Advisor suggests cheaper alternatives. CSV import, spend by model/task, 14-day trend. Runs 100% locally — no sign-up, no API keys.