Multi-tenancy in Agentic AI – session recap

Software as a Service (SaaS) enables software development companies (SDCs) to deliver their products as managed services to customers. Whether offered business to business (B2B) or business to consumer (B2C), the way a product—new or existing—is designed to support multiple tenants (such as customers, user groups, or internal departments) can vary based on many factors. ‑to‑business (B2B) or business‑to‑consumer (B2C), the way a product—new or existing—is designed to support multiple tenants (such as customers, user groups, or internal departments) can vary based on many factors.

Multi-tenancy on Microsoft Learn - SaaS and Multitenant Solution Architecture - Azure Architecture Center | Microsoft Learn

To help you with architectural considerations (questions like “how will you define tenant”, “how will you deploy your infrastructure to support multitenancy”, “how will you monitor, automate, scale”, etc.), we have an assessment (available here) to guide you step by step through that. We’ve even built a checklist (more here) as a starting point to help you design and build your multitenant solution with ease.

In general, most common ways / perspectives we can handle multi-tenancy are:

1. Application (do we have shared apps for all customers, or we create apps specific for customers, …)
2. Data (do we use one data storage, do we multiple, how do we effectively query through data, how do we scale / partition, when do we scale …)
3. Infrastructure (do we handle one region, do we handle multiple, how do we do multiple customers across multiple regions, …)
4. Agents (how we align agent with tenants, do we create agent per tenant, or load tenant-based tools, prompts, data in shared ecosystem, …)

In the session (video available here, slides here, demos here) we were focusing on 2 things (with concrete examples):

1. Application and data perspective with proposed architecture patterns (stamp pattern, circuit breaker)
2. Agent usage with different techniques and patterns (semantic caching, agent id)

Starting with a clean multi-tenant design is ideal, but we recognize that this is often not the reality in the field. With that in mind, the demos we are presenting in the session are built with:

1. having web (UI) and API (backend business logic and agent usage) applications, which are not multi-tenant, and then go step by step and transform, highlighting what it takes to change the app and backend from single to multi customer usage.
2. Continuing with focusing on how to handle agent usage with different approaches (semantic caching, loading tools / prompts based on tenant identifier, etc.) to server multiple tenants efficiently.

Recording of the session

After you watch the video, you’ll notice that multi-tenancy comes with many considerations, challenges, and different ways to approach it, locally and on Azure.

Azure—and our team—are here to simplify this journey for you and your customers.

p.s.

Check also other webinars and recordings from our SDC academy.