// the take
A .NET 10 + React 19 scaffolding kit for multi-tenant SaaS — auth, billing, multitenancy, real-time, file storage, webhooks, and background jobs wired up before you write your first line of business logic. You run `fsh new MyApp` and get detached source with real project references, not a NuGet black box. Aimed at solo devs or small teams who want to skip the plumbing and ship a product.
- The one-shot DbMigrator pattern is the right call — migrations never run at API startup, which prevents the class of production incidents where a multi-instance deploy races on schema changes.
- Module boundaries enforced by NetArchTest architecture tests, not just folder convention. If you accidentally reach across a module boundary, CI breaks. That discipline is rare in starter kits.
- Source-generated CQRS via Martinothamar's Mediator instead of MediatR — no reflection overhead at runtime, and the source generator surfaces wiring errors at compile time rather than startup.
- The `.agents/` directory ships with rules, skills, and workflows for AI coding agents (separate files for CLAUDE.md, AGENTS.md, GEMINI.md). First starter kit I've seen that treats AI pair programming as a first-class workflow.
- The 'it's all yours' pitch cuts both ways: once you scaffold, upstream fixes don't merge back. You own the Chat module, the Billing module, and the Webhooks module forever — including security patches in code you didn't write and probably don't fully understand yet.
- Admin and dashboard are two entirely separate React apps with duplicated shadcn components. When Tailwind v4 or Radix ships a breaking change, you fix it twice, in two different package-lock files, hoping you didn't miss anything.
- Finbuckle multitenancy adds non-obvious complexity that the README treats as a solved problem. Per-tenant migrations, tenant resolution middleware, and isolated DbContexts interact in ways that will surprise you the first time a tenant provisioning step partially fails.
- The breadth is also the adoption risk: you inherit Hangfire, SignalR, a Billing module, a Chat module, and a Tickets module on day one. If your MVP needs none of those, you're now maintaining dead code and explaining to new contributors why it's there.
We read the README, the directory tree and a key source file in June 2026. Nobody pays to be listed. How we judge →