Structuring an AI SaaS with FastAPI and Next.js: Key Architecture Decisions

An AI SaaS differs from ordinary software because each request carries real cost, responses take seconds, provider dependencies are outside the team's control, and customers supply sensitive data. The guide recommends using Next.js for rendering and interface work while keeping business logic, data access and all model calls in a FastAPI backend, which also yields a typed API contract. A central rule is that browsers never contact LLM providers or hold provider keys, and authentication should use short-lived tokens with a rotating refresh token in an HttpOnly cookie. The article also advises choosing a tenancy model early, enforcing tenant isolation in a single data-access layer, and routing AI calls through an internal gateway that manages credentials, model allow-lists, retries, usage accounting and quotas. Long-running tasks such as document processing should be moved off the request cycle to avoid proxy timeouts.