AI agent glossary
Multi-tenancy
Serving many separate customers from one agent system while guaranteeing that each customer's data, memory, and credentials stay isolated from the others.
Once an agent serves more than one customer, every place data flows becomes a place it can leak: retrieval that searches all documents, memory saved without a customer key, a cache keyed only by the question, a tool that accepts any account ID.
The central rule is that tenant identity comes from your authentication layer, never from the model. Tools receive the tenant from trusted code, so no prompt can talk the agent into another customer's data. Enforcing isolation in the database itself, for example with row-level security, adds a second line of defense.
Multi-tenancy also covers fairness: per-tenant rate limits and budgets keep one heavy customer from slowing everyone else. Test isolation the way an attacker would.
A support agent's search tool adds the signed-in customer's tenant ID to every query in code, so even a prompt asking for 'all invoices' only ever returns that customer's documents.
Learn more in the course
Related terms
Go from definitions to doing
Learn to use and build AI agents in 130 practical lessons.