Show HN: Replay – Audit silent prompt cache misses in AI agent transcriptshttps://replay.doctor
For the systems engineers looking at the underlying architecture of Replay, here is exactly how we model the plumbing and why we chose these specific design constraints: 1. Why Go Instead of Python? We built this strictly as a zero-dependency local CLI binary. Forcing an engineering lead or security auditor to handle Python virtual environments, pip dependency drift, or heavy Docker runtimes completely ruins the frictionless developer experience. 2. How We Model Server-Side State Machines: Anthropic and OpenAI keep their eviction layers in a black box. To build our decay maps, we forensic…
