Vault — privacy layer for AI agents Building with Vapi and kept hitting the same problem: sensitive user data (names, numbers, emails) going straight to the LLM verbatim. Built Vault to fix it — masks PII before the prompt leaves your server, sends clean tokens to the LLM, restores real values before your agent executes. The model never sees raw data. → https://ycsummer.vercel.app/ Early access is open, would love feedback from anyone building voice agents with real user data.