This is the telemetry NGOS runs on its own AI activity — what each model costs, where every dollar goes, and the step-by-step trace of any run. Pick a window; the figures recompute.
Spend breakdown
Recent runs tap to trace
How to read this. Cost is real — computed from published per-token prices (Opus $5/$25, Sonnet $3/$15, Haiku $1/$5, economy $0.20/$0.40 per 1M). Every call is tagged with the model, the task, and the gate it passed: draft (a human approves before use), human-gated (a person decides), or auto (non-employment, no decision required). Latency figures are representative. The "saved vs. all-premium" number is what routing kept off the bill versus running every task on the top model.