The problem
Every indie AI project I started repeated the same week of setup: auth, payments, rate limiting, a usable LLM streaming layer, an email transactional pipeline, and the tooling to deploy it. By the time the boilerplate was done, the urge to ship was gone.
The approach
OpenKit is the boilerplate I wish I'd had. The constraint was sharp: opinionated where opinions compound, swappable where they don't. The rules:
- Every primitive must be production-runnable on day one. No “wire this up later” TODOs.
- Type-safety from the database to the client. Drizzle, tRPC, Zod, no escape hatches.
- The first commit should already deploy. If it doesn't, the boilerplate has failed.
What's inside
Auth
Better Auth with email + OAuth, session rotation, and org support out of the box.
Billing
Stripe checkout + customer portal + webhook handlers, usage metering scaffolded.
AI
Streaming chat, server-side tools, structured outputs. Model-agnostic.
React Email templates with a dev mailbox preview and Resend in prod.
What I learned
The hardest part of a boilerplate is knowing what to leave out. Every additional abstraction costs the user a mental model. The version that shipped is half the size of the first draft, and a lot more useful for it.