Thanks, Adam. I agree; the database filter is the easy part. The harder part is anything that lives between requests, like a summary or a cached plan made for one user and then reused for the next. In MemorySync, every call has to name the end user, and a search only looks at that user's memories. That helps, but it can't protect a summary your agent keeps in its own working memory. Have you seen this happen in a real system, or mostly in testing?