Technical Stack · Used at Richbay

Low-operations AI product prototype

Build and release a small AI product without committing early to a large application platform or a long list of services.

5 component roles3 explicit defaultsReviewed 2026-09-05
← All StacksInspect the defaults

Components and roles

Every component earns its place.

  1. 01
    Next.js application

    Provides the public interface, static content, route handlers, and server-rendered paths when required.

  2. 02
    Cloudflare Workers deployment path

    Runs the application at the edge through the current compatible Next.js adapter.

  3. 03
    Server-side model boundary

    Keeps provider credentials private and applies validation, timeouts, budgets, and safe failure behavior.

  4. 04
    Optional D1 and R2 bindings

    Add structured records or generated artifacts only after the product flow proves the need.

  5. 05
    Release and observation loop

    Connects each deployment to a commit, route checks, error visibility, cost signals, and a rollback rule.

Data flow

Browser → validated application route → model provider when needed → optional D1/R2 record → user-visible result → release and error review

Default choices

Start here, then switch for a stated reason.

Next.js on Workers

vinext for a compatible new app; OpenNext for an existing app with a migration gap

Selection reason
Matches Cloudflare’s current guidance while avoiding an unverified production migration.
Switch when
Re-evaluate the adapter when compatibility, maintenance status, or application requirements materially change.

Structured persistence

No database until the flow needs durable records; D1 as the Cloudflare-native starting point

Selection reason
Avoids operating unused state while keeping a direct binding path when persistence becomes real.
Switch when
Use an existing Postgres system or another store when relational features, portability, regions, or team operations make it the better fit.

Artifact storage

No object store until files or immutable exports exist; R2 when they do

Selection reason
Keeps the first release small and separates large objects from structured records.
Switch when
Use an existing object store when data residency, tooling, or organizational standards outweigh consolidation.

Operating assumptions

Cost, privacy, alternatives, and limits.

Cost assumption
Begin with static and request-driven work. Model inference, Workers usage, storage, logs, and human review all scale separately and need explicit budgets.
Privacy boundary
Only validated task data should cross the server-side model boundary. Secrets stay in deployment configuration; durable data needs a retention and deletion owner.
Reference implementation
Richbay currently runs an existing Next.js 16 application on Cloudflare Workers through OpenNext; new applications should evaluate Cloudflare’s recommended vinext path first.
Alternatives
Use a static export when the product has no server requirements; use another host or application framework when its runtime and operating model fit the team better.
Known limitation
This stack supports a low-operations prototype. It does not by itself provide production security, model evaluation, compliance, data recovery, or product-market fit.

Human checkpoints

Keep accountable decisions visible.

  1. Approve the one-flow scope and data boundary
  2. Review production configuration and the exact release artifact
  3. Verify the public version and own rollback

Evidence and sources

See what supports this record.

Used at Richbay
Last reviewed 2026-09-05. Recheck current provider documentation, limits, pricing, and terms before adoption.