Ship an AI SaaS in an Afternoon

Most of the work in launching an AI product has nothing to do with AI. It is sign-up flows, password resets, a billing integration that survives a failed webhook, and a database schema you will not regret in three months. The starter exists so you can skip that part.

What you get

Clone the repo and you have a Next.js 16 app with:

• Better-Auth for email/password and Google sign-in, with email verification and password reset already wired to Resend.
• Drizzle ORM on Postgres, with typed schemas and a CLI migration workflow.
• Stripe subscriptions across three tiers, including Checkout, the customer portal, and a webhook handler that keeps plan state in your own database.
• A streaming chat layer on the Vercel AI SDK, with persisted conversations and a model picker.
• Task templates for one-shot generation, including one that returns validated JSON.
• Usage limits enforced per plan before a request reaches the model.

The parts you actually change

Three files carry most of your product's personality:

config/ai.ts holds the system prompt, the history window, and the sampling temperature. This is where your assistant stops sounding generic.

config/tasks.ts defines the task templates. Each one is a slug, a few form inputs, and a system prompt — add an entry and a new page appears under /dashboard/tasks.

config/subscriptions.ts defines your tiers, prices, and per-plan request limits. The limits here are enforced server-side by checkRateLimits(), so raising a number in this file raises the real cap.

Deploying

The app runs on Vercel with no special configuration. Point DATABASE_URL at a hosted Postgres instance, add your API keys, run npm run db:migrate, and deploy. The Stripe webhook needs one endpoint pointing at /api/webhooks/stripe — everything else is environment variables.

The interesting work starts after that: what your product does that nobody else's does.