Built Authorizer(https://authorizer.dev) — open source auth with 2K GitHub stars and 200+ Railway deployments. Now shipping RAGAuth.tech (https://ragauth.tech) on top of it. The problem I kept seeing: teams add AI to their product, point it at their documents, and assume permissions are inherited. They're not. The AI finds the most relevant answer — from everyone's data. An engineer asking "how do raises work?" gets the compensation memo. Not because of a bug. Because vector search has no concept of need-to-know. RAGAuth fixes this at the retrieval layer. Permissions checked before the search, not after. Just launched ragauth.tech — managed cloud waitlist open, founding pricing at $149/mo. Looking for feedback from anyone building multi-tenant AI or RAG pipelines. What permission problems are you hitting?