The production web app is a TanStack Start application built with the Cloudflare Vite plugin. Its Worker handles SSR, Better Auth, oRPC, the files gateway, MCP, AI streaming, OG images and cron. Expo calls the same HTTP API. There is no separate API deployment.
Runtime and infrastructure
| Component | Responsibility |
|---|---|
| TanStack Start | Routes, SSR, hydration and request handlers |
| Worker fetch | Carries Cloudflare waitUntil into the request runtime before dispatch |
| Hyperdrive | Pools connections to Postgres; query caching must stay disabled |
| Postgres, hosted on Neon | Users, Organizations, domain records, leases, outboxes and approvals |
| R2 | Private objects, accessed through the files-sdk fetch adapter |
| Worker scheduled | Runs the six jobs in one scoped Effect program |
The default build targets Cloudflare. DEPLOY_TARGET=node selects Nitro for a Node deployment. vite dev always uses Nitro, with a shim for cloudflare:workers and direct DATABASE_URL access. The Node path is useful locally; the built Worker remains the compatibility test for production.
Read the implementation
- Request lifecycle follows one authenticated request.
- Client/server boundary explains which imports are safe.
- Effect at the edges covers deadlines, cancellation and retries.
- Observability distinguishes application logs from Worker traces.
Shared code belongs in a package when web and mobile need the same contract or behavior. The package reference lists actual exports. Adding a feature usually extends the existing Worker, schema and router, rather than provisioning another service.