Ask Me Bot Case Study # 1: Bot-Level Content Isolation

The problem here was that multiple bots on the same account were sharing a vector pool with only an application-layer filter separating them. The fix replaced that soft boundary with per-bot vector namespaces at the database level, meaning cross-bot data bleed became structurally impossible rather than just unlikely. It also introduced an assignment-first ingestion flow that eliminated the window where unassigned documents were temporarily queryable, plus synchronous cache invalidation on every assignment change and an audit log so teams could see exactly what each bot knew and when that changed.