Replay

Replay is a local CLI tool that detects when AI providers silently re-bill your conversation history due to broken prompt cache. It forensically analyzes on-disk log transcripts from Claude Code, Cursor, Aider and other AI agents, replaying them against provider caching logic to identify exact token waste.

Built as a zero-telemetry, zero-dependency Go application, Replay helps teams understand unexpected AI infrastructure cost spikes by pinpointing cache misses before they compound across agent runs.