LLMProxy ships WAF, cache, plugin stack

LLMProxy is aiming at the messy middle between proxy, WAF, and observability, which is exactly where LLM teams keep rebuilding the same controls.

• –The ring-based pipeline makes policy stages explicit, so auth, masking, routing, and post-processing can be reasoned about separately instead of buried in middleware.
• –WASM-sandboxed plugins are the right extensibility bet: they reduce blast radius and make third-party policy logic safer to run.
• –Combining cache and WAF controls in one layer should help with both latency and spend, especially for repetitive or high-volume LLM traffic.
• –The SOC UI and per-plugin metrics are useful if teams actually want to debug prompt attacks, refusals, and quality regressions in real time.
• –The hard part will be proving the security story under adversarial traffic while keeping the plugin system simple enough for outsiders to adopt.