Shipping on night one

Every new product starts the same way. You have an idea at 9pm, you run pnpm create, and by midnight you are not building the idea — you are reading a migration guide about session cookies.

The plumbing is not hard. It is just long. Auth, a database, a billing loop, transactional email, a marketing page that does not embarrass you, an analytics pixel, an error reporter. None of it is interesting, all of it is required, and it is almost identical in every product you will ever build.

The cost of starting from zero

Here is the honest accounting for a solo developer, from empty directory to something a stranger could pay for:

Piece	Realistic first pass
Auth + session handling	1–2 days
Database schema + access rules	1 day
Stripe checkout, portal, webhooks	2–3 days
Marketing site with real SEO	2 days
Email, analytics, error tracking	1 day
Deployment and environments	1 day

That is over a week before the first line of product code. Worse, it is a week you will spend again on the next project, and you will make slightly different decisions each time, so nothing you learn compounds.

What "done" should mean

A starter earns its keep when every integration is either finished or honestly labelled. The rule we hold this template to:

If a key is missing, the feature degrades. It never crashes, and it never pretends to work.

Concretely, that means a fresh clone with only Convex and Clerk configured still builds, still prerenders every marketing page, and still renders the pricing page — the upgrade button just tells you Stripe is not configured yet.

// convex/stripe.ts
function getStripe(): Stripe | null {
  const key = process.env.STRIPE_SECRET_KEY
  if (!key) return null
  return new Stripe(key, { apiVersion: '2025-10-29.clover' })
}

Three lines, and the difference between a template you can run and a template you have to finish before you can run it.

Pick a stack and stop relitigating it

The second thing a starter buys you is the end of the debate. This one is settled:

• TanStack Start for the app — routing, data, forms and the server runtime from one family.
• Convex as the only database. There is no second store to keep in sync.
• Cloudflare Workers as the host. It serves the app; it is not a backend.
• Clerk for identity, so the same accounts work when a native client shows up later.
• Stripe for subscriptions on the web.

You can disagree with any of these. What you should not do is rediscover the tradeoffs at 11pm on a Tuesday while you are trying to validate an idea.

The part that is actually yours

Strip out the plumbing and what remains is small: a few screens, a data model, and the specific thing you believe about the problem. That is the part worth your first evening.

Everything else is in this repository already.