Connections

A Prisma.Connection is a credential (an API key) for one database. The name prop is optional — alchemy derives one from the logical ID and appends the resource’s instance identity so connections stay unambiguous — and every secret output is Redacted so nothing prints in logs or plan output.

import * as Prisma from "alchemy/Prisma";

const connection = yield* Prisma.Connection("api", {
  database: postgres,
});

Connection outputs

connection.databaseUrl; //                conventional app URL (pooled → direct → Accelerate)
connection.directConnectionString; //     direct Postgres URL
connection.pooledConnectionString; //     pooled Postgres URL
connection.accelerateConnectionString; // Accelerate URL, when the database has one
connection.origin; //                     parsed direct connection components
connection.pooledOrigin; //               parsed pooled connection components

databaseUrl is the serverless-safe default for application traffic: it prefers the pooled endpoint, then direct, then Accelerate. For a standalone database under a project created with createDatabase: false, pass it (and the direct URL for migration tooling) straight into an app’s env:

const app = yield* Prisma.Compute("api", {
  project,
  path: "./app",
  env: {
    DATABASE_URL: connection.databaseUrl,
    DIRECT_URL: connection.directConnectionString,
  },
});

If Prisma.Project created the project’s default database, omit these env entries. Prisma injects system-managed DATABASE_URL and DATABASE_URL_POOLED values into Compute deployments automatically.

The origin shape

origin and pooledOrigin are the same connections parsed into the structured shape Postgres consumers like Cloudflare.Hyperdrive accept — the same shape Neon branches and PlanetScale roles materialize:

type PostgresOrigin = {
  scheme: "postgres" | "postgresql";
  host: string;
  port: number;
  database: string;
  user: string;
  password: Redacted.Redacted<string>;
};

Hyperdrive is itself a pooler, so hand it the direct origin:

const hyperdrive = yield* Cloudflare.Hyperdrive.Connection("api-hd", {
  origin: connection.origin.as<Prisma.PostgresOrigin>(),
});

(The .as<...>() narrows the output: a connection to a legacy Accelerate-only database has no direct Postgres endpoint, so origin is typed optional.)

The Connect binding

Inside a Prisma Compute app, AWS Lambda function, Cloudflare Worker, or Cloudflare Container, Prisma.Connect resolves the bound connection at runtime as a typed client:

Effect.gen(function* () {
  const db = yield* Prisma.Connect(connection);
  const sql = yield* SQL.Postgres({ url: db.databaseUrl });

  return {
    fetch: Effect.gen(function* () {
      const users = yield* sql`SELECT * FROM users`;
      return yield* HttpServerResponse.json(users);
    }),
  };
}).pipe(Effect.provide(Prisma.ConnectBinding));

At deploy time the binding carries the connection’s outputs into the host’s environment (env vars on Compute, Lambda, and Containers; secret text bindings on Workers); at runtime db.databaseUrl and friends read them back as Redacted values.

A container is a real process with no workerd bindings, so env vars are its only channel — which is exactly why Prisma.Connect reaches it and Cloudflare.Hyperdrive.Connect does not. Start the container with Cloudflare.Containers.layer(Api, { enableInternet: true }) so it can reach the database. See Bind a capability into an effectful container.

Rotation

Change rotate from false to true to mint fresh credentials on the next deploy while keeping the connection’s identity:

const connection = yield* Prisma.Connection("api", {
  database: postgres,
  rotate: true,
});

Prisma attempts to revoke the previous credentials on a best-effort basis. Downstream consumers receive the new secret on the same deploy through their bindings. To rotate again, deploy once with rotate: false, then change it back to true.

Where next

• Connect from Workers — both wiring styles for Cloudflare, side by side.
• Hyperdrive — the full pooling walkthrough.

Reference:

• Connection API reference