Tokenwatch

Track what language-model calls cost and how fast they are: spend and latency overview by model and feature, a per-model grid with error and cache rates, team budgets with forecast, and an alert feed for spikes.