Shunt is a reverse proxy for LLM requests that detects recursive agent loops and enforces spend caps. It is built for teams sending model traffic through an OpenAI-compatible gateway.
It also says credentials stay in memory only during the request and are not written to disk or logs.