The SvelteKit + Postgres B2B starter — the boring 80% already wired.

Built for AI coding agents

The kit ships AGENTS.md, a generated API reference andllms.txt, so Claude Code, Cursor and Copilot start with the file map and the public API already in context instead of inferring them.

They are included because they cost nothing to ship and make the kit more usable — not because they are unusual. The Svelte CLI'snpx sv add ai-tools already scaffolds a framework-levelAGENTS.md into any SvelteKit project; these are the product-specific files layered on top of that baseline.

• AGENTS.md — architecture decisions, command shortcuts, and "things that will bite you" so an agent starts productive, not fumbling.
• TypeDoc API reference — generated from the code, a machine-readable spec of every export that cannot silently rot.
• A public docs portal + llms.txt — deep-dive guides and a site index an agent can retrieve, not just a human can read.
• 454 automated tests — the trust layer. AI code without tests is a liability. AI code with 454 tests is a foundation.

Not a bolt-on: the full agent context lives in the repo from day one.Read the AGENTS.md.

How it compares

• All three kits share the same tested core — orgs, invites, RBAC, seat billing, audit log. The difference is the data layer.
• If you are already on Supabase and want managed Auth/RLS/realtime, see the SvelteKit + Supabase Starter ($99 early-bird).
• If you want zero-ops on Cloudflare with SQLite/D1, see the Multi-tenant SvelteKit Starter ($79 early-bird, 475 tests).
• Read the full comparison: Postgres vs Supabase · Drizzle vs Prisma

Who this is for

This page is written for one reader: someone who has already chosen SvelteKit, is building a B2B product that will hold more than one organisation, and has run out of patience for the tenancy layer.

• Teams are on your roadmap, not in your codebase.The next milestone is orgs, workspaces or accounts — the point where "just add a tenant_id column" stops being true and every query starts being somebody's data leak.
• A demo date is closer than the feature is.The work in this kit is the work you cannot show on Friday: an invite that expires before it is claimed, two members claiming the same invite, a join that lands after the seat limit is reached.
• You have already decided on Postgres.If you want managed Auth on Supabase instead, that isthe other kit; if you want zero-ops SQLite on Cloudflare D1, that isthis one.
• Your next customer has more than one team in it.When the first real account asks who changed what and when, the audit log and the server-side RBAC are the two answers you would otherwise be writing from scratch — under a deadline, in someone else's code.

What you're buying instead of building it

The honest version of "$79 or do it myself" is not a number of hours — you know your rate better than we do. It is the list of what you would otherwise have to write, and the list of what you would only notice was missing after a customer hit it.

The tests are the part that earns the money. They are the written record of the edge cases: an invite link that expired, two members claiming the same invite, a join arriving after the seat limit, a member's forged POST trying to promote themselves, an audit row with no edit path. None of that shows up in a screenshot, and all of it is what a paying account finds first.

If your own estimate for that list is a few weeks, then $79 is not really the decision — the weeks are. We don't publish an hour count, because we cannot measure yours, and a number we made up would only be an ad.

Dive deeper