Todo API Walkthrough

The project includes an executable Bun API example. It combines Services, Layers, Effects, Runtime, Scope, better-result, Bun SQL and the ITI adapter. Use it as a reference for organizing a real application; in your own project, the example files can live wherever your application structure calls for them.

Dependency graph

AuthService ──────────────┐
  ├── UserRepository ─────┤
  ├── SessionRepository ──┤── Database (Layer.scoped)
  └── PasswordHasher       │
TodoService ──────────────┘
  └── TodoRepository ─────┘

The Services contain behavior and yield the dependencies they need. The Layer file is the composition root:

const DatabaseLive = Layer.scoped(
  Database,
  async () => {
    const database = new Database(new SQL(':memory:', { adapter: 'sqlite' }))
    await database.initialize()
    return database
  },
  (database) => database.close()
)

const RepositoriesLive = Layer.merge(
  Layer.make(UserRepository),
  Layer.make(SessionRepository),
  Layer.make(TodoRepository)
)

const ServicesLive = Layer.merge(
  Layer.make(PasswordHasher),
  Layer.make(AuthService),
  Layer.make(TodoService)
)

export const AppLive = Layer.merge(DatabaseLive, RepositoriesLive, ServicesLive)

Request lifecycle

Each HTTP handler runs a program in the shared Runtime:

const handleRequest = Effect.fn(async function* () {
  const { userId } = yield* Result.await(requireUser(request))
  const todos = yield* TodoService
  const items = yield* Result.await(todos.list(userId))

  return Result.ok(items)
})

const result = await runtime.run(handleRequest)

The request gets an execution child Scope. The SQLite client belongs to the root Scope and remains available across requests. A request-local file, lease or stream can use Effect.acquireRelease and be cleaned up as soon as that handler settles.

Run a project based on this example

bun run src/index.ts

The server listens on http://localhost:3333 and seeds demo@example.com / demo1234 in the reference implementation. In a new application, use your own entry point and configuration. The important part is that the entry point passes the actual AppLive Layer and typed main Program to NodeRuntime.runMain. That one boundary owns the application Runtime, passes its resolver context to request handlers, stops the server when CurrentAbortSignal aborts, and disposes the Layer resources.

Why this shape scales

• The HTTP layer knows only the Runtime and domain Results.
• Services express behavior and dependencies without ITI imports.
• Layers make production and test environments explicit.
• Runtime owns graceful shutdown.
• better-result keeps error unions typed from database to HTTP response.