Is Refract or SupaStarter better in 2026?

Neither wins universally. Choose SupaStarter if a Next-first (or Nuxt/TanStack) surface, broad prebuilt kit, and familiar conventions dominate your constraints. Choose Refract if you expect packaging experiments, webhook correctness, vendor swaps, and AI-assisted changes to land cleaner when boundaries are already paid for.

How do Refract and SupaStarter differ when AI writes most of the code?

Agents work in competent codebases on both. The question is how much architecture they must invent. Refract leans on explicit contracts (tools.*, GraphQL schema, AllScopes, AGENTS.md, tests) so agents have somewhere logic belongs and a feedback loop. SupaStarter leans on Next-shaped conventions and a wide surface agents already know from the ecosystem. Neither claim is 'better codegen magic'.

How do Refract and SupaStarter differ on billing architecture?

Refract separates products (packaging/scopes) from prices (commercial terms), routes Stripe only through tools.paymentProcessor, and reconciles Stripe into Postgres via a queued stripeWebhookConsumer with processed_webhooks_events idempotency. SupaStarter publicly emphasizes multi-provider billing (Stripe and several merchant-of-record options) inside a Next.js monorepo. That breadth is a real win; depth of ownership still needs a source audit.

Does SupaStarter lock you into Next.js?

Publicly, SupaStarter is Next.js-first and also offers Nuxt and TanStack Start kits. The product surface is still framework-shaped. Refract is Express + Apollo GraphQL with a React/Vite portal and Astro marketing app, so the domain layer is not hosted inside a Next request lifecycle.

What does RBAC as a layer mean in Refract?

Permissions are AllScopes strings resolved from ProductRoleScope rows for the org's active subscription product and membership role, with Redis caching and hasScopes GraphQL middleware. Checks ask for capabilities like members:invite or billing:upgrade:subscriptions, not only role === admin.

Who should pick SupaStarter over Refract?

Founders and agencies who want a Next-first (or Nuxt) product surface, stronger out-of-the-box marketing/i18n/admin coverage, familiar ecosystem conventions, and multiple payment providers marketed as ready. Also teams whose primary risk is still market fit rather than billing rewrite tax.

Refract vs. SupaStarter

Convention-driven Next starters vs a contract-driven Express/GraphQL foundation, in a year when AI writes most of the code.

Yoann Jaspar

October 21, 2026 13 min read Evaluating Refract

Pick SupaStarter when you want a Next-first (or sibling-kit) product surface, broad prebuilt auth/marketing/i18n/admin coverage, and familiar ecosystem conventions.
Pick Refract when you want contract-first boundaries: Express + Apollo GraphQL, scope-based RBAC tied to products, Stripe behind tools.paymentProcessor, queues that already prove BullMQ and SQS.
Both let coding agents write most of the code. Refract is designed so agents invent less architecture: where logic belongs, which interface is allowed, which pattern to copy, which test proves done.
Billing and RBAC still matter. Refract splits product vs price in Postgres, reconciles Stripe through a queued idempotent consumer, and maps AllScopes via ProductRoleScope. SupaStarter's public multi-provider billing breadth is a real brochure win.
No universal winner. Choose the failure mode you are willing to own for years of human and agent edits.

FAQ

Is Refract or SupaStarter better in 2026?: Neither wins universally. Choose SupaStarter if a Next-first (or Nuxt/TanStack) surface, broad prebuilt kit, and familiar conventions dominate your constraints. Choose Refract if you expect packaging experiments, webhook correctness, vendor swaps, and AI-assisted changes to land cleaner when boundaries are already paid for.
How do Refract and SupaStarter differ when AI writes most of the code?: Agents work in competent codebases on both. The question is how much architecture they must invent. Refract leans on explicit contracts (tools.*, GraphQL schema, AllScopes, AGENTS.md, tests) so agents have somewhere logic belongs and a feedback loop. SupaStarter leans on Next-shaped conventions and a wide surface agents already know from the ecosystem. Neither claim is 'better codegen magic'.
How do Refract and SupaStarter differ on billing architecture?: Refract separates products (packaging/scopes) from prices (commercial terms), routes Stripe only through tools.paymentProcessor, and reconciles Stripe into Postgres via a queued stripeWebhookConsumer with processed_webhooks_events idempotency. SupaStarter publicly emphasizes multi-provider billing (Stripe and several merchant-of-record options) inside a Next.js monorepo. That breadth is a real win; depth of ownership still needs a source audit.
Does SupaStarter lock you into Next.js?: Publicly, SupaStarter is Next.js-first and also offers Nuxt and TanStack Start kits. The product surface is still framework-shaped. Refract is Express + Apollo GraphQL with a React/Vite portal and Astro marketing app, so the domain layer is not hosted inside a Next request lifecycle.
What does RBAC as a layer mean in Refract?: Permissions are AllScopes strings resolved from ProductRoleScope rows for the org's active subscription product and membership role, with Redis caching and hasScopes GraphQL middleware. Checks ask for capabilities like members:invite or billing:upgrade:subscriptions, not only role === admin.
Who should pick SupaStarter over Refract?: Founders and agencies who want a Next-first (or Nuxt) product surface, stronger out-of-the-box marketing/i18n/admin coverage, familiar ecosystem conventions, and multiple payment providers marketed as ready. Also teams whose primary risk is still market fit rather than billing rewrite tax.