I built this to demonstrate the multi-tenancy patterns required for production AI SaaS products. Each tenant's data is isolated at the database level, with per-organization configuration controlling which tools and capabilities are enabled for their users.

Conversations are scoped per user per tenant. When a thread exceeds 8,000 tokens, a BullMQ background worker automatically summarizes it and replaces the context window, keeping costs under control without losing conversational continuity.

The system uses the same Express 5, PostgreSQL, Redis, and BullMQ monorepo structure as the other portfolio apps, making the multi-tenancy layer the clear focus. It's a blueprint for any AI product that needs to serve multiple organizations from a single deployment.